Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2017-03-16' into staging
[qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
44 #include "qemu/id.h"
45 #include "qapi/util.h"
46
47 #ifdef CONFIG_BSD
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
50 #ifndef __DragonFly__
51 #include <sys/disk.h>
52 #endif
53 #endif
54
55 #ifdef _WIN32
56 #include <windows.h>
57 #endif
58
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60
61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
63
64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
66
67 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
68 QLIST_HEAD_INITIALIZER(bdrv_drivers);
69
70 static BlockDriverState *bdrv_open_inherit(const char *filename,
71 const char *reference,
72 QDict *options, int flags,
73 BlockDriverState *parent,
74 const BdrvChildRole *child_role,
75 Error **errp);
76
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist;
79
80 #ifdef _WIN32
81 static int is_windows_drive_prefix(const char *filename)
82 {
83 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
84 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
85 filename[1] == ':');
86 }
87
88 int is_windows_drive(const char *filename)
89 {
90 if (is_windows_drive_prefix(filename) &&
91 filename[2] == '\0')
92 return 1;
93 if (strstart(filename, "\\\\.\\", NULL) ||
94 strstart(filename, "//./", NULL))
95 return 1;
96 return 0;
97 }
98 #endif
99
100 size_t bdrv_opt_mem_align(BlockDriverState *bs)
101 {
102 if (!bs || !bs->drv) {
103 /* page size or 4k (hdd sector size) should be on the safe side */
104 return MAX(4096, getpagesize());
105 }
106
107 return bs->bl.opt_mem_alignment;
108 }
109
110 size_t bdrv_min_mem_align(BlockDriverState *bs)
111 {
112 if (!bs || !bs->drv) {
113 /* page size or 4k (hdd sector size) should be on the safe side */
114 return MAX(4096, getpagesize());
115 }
116
117 return bs->bl.min_mem_alignment;
118 }
119
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path)
122 {
123 const char *p;
124
125 #ifdef _WIN32
126 if (is_windows_drive(path) ||
127 is_windows_drive_prefix(path)) {
128 return 0;
129 }
130 p = path + strcspn(path, ":/\\");
131 #else
132 p = path + strcspn(path, ":/");
133 #endif
134
135 return *p == ':';
136 }
137
138 int path_is_absolute(const char *path)
139 {
140 #ifdef _WIN32
141 /* specific case for names like: "\\.\d:" */
142 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
143 return 1;
144 }
145 return (*path == '/' || *path == '\\');
146 #else
147 return (*path == '/');
148 #endif
149 }
150
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152 path to it by considering it is relative to base_path. URL are
153 supported. */
154 void path_combine(char *dest, int dest_size,
155 const char *base_path,
156 const char *filename)
157 {
158 const char *p, *p1;
159 int len;
160
161 if (dest_size <= 0)
162 return;
163 if (path_is_absolute(filename)) {
164 pstrcpy(dest, dest_size, filename);
165 } else {
166 p = strchr(base_path, ':');
167 if (p)
168 p++;
169 else
170 p = base_path;
171 p1 = strrchr(base_path, '/');
172 #ifdef _WIN32
173 {
174 const char *p2;
175 p2 = strrchr(base_path, '\\');
176 if (!p1 || p2 > p1)
177 p1 = p2;
178 }
179 #endif
180 if (p1)
181 p1++;
182 else
183 p1 = base_path;
184 if (p1 > p)
185 p = p1;
186 len = p - base_path;
187 if (len > dest_size - 1)
188 len = dest_size - 1;
189 memcpy(dest, base_path, len);
190 dest[len] = '\0';
191 pstrcat(dest, dest_size, filename);
192 }
193 }
194
195 void bdrv_get_full_backing_filename_from_filename(const char *backed,
196 const char *backing,
197 char *dest, size_t sz,
198 Error **errp)
199 {
200 if (backing[0] == '\0' || path_has_protocol(backing) ||
201 path_is_absolute(backing))
202 {
203 pstrcpy(dest, sz, backing);
204 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
205 error_setg(errp, "Cannot use relative backing file names for '%s'",
206 backed);
207 } else {
208 path_combine(dest, sz, backed, backing);
209 }
210 }
211
212 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
213 Error **errp)
214 {
215 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
216
217 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
218 dest, sz, errp);
219 }
220
221 void bdrv_register(BlockDriver *bdrv)
222 {
223 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
224 }
225
226 BlockDriverState *bdrv_new(void)
227 {
228 BlockDriverState *bs;
229 int i;
230
231 bs = g_new0(BlockDriverState, 1);
232 QLIST_INIT(&bs->dirty_bitmaps);
233 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
234 QLIST_INIT(&bs->op_blockers[i]);
235 }
236 notifier_with_return_list_init(&bs->before_write_notifiers);
237 bs->refcnt = 1;
238 bs->aio_context = qemu_get_aio_context();
239
240 qemu_co_queue_init(&bs->flush_queue);
241
242 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
243
244 return bs;
245 }
246
247 static BlockDriver *bdrv_do_find_format(const char *format_name)
248 {
249 BlockDriver *drv1;
250
251 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
252 if (!strcmp(drv1->format_name, format_name)) {
253 return drv1;
254 }
255 }
256
257 return NULL;
258 }
259
260 BlockDriver *bdrv_find_format(const char *format_name)
261 {
262 BlockDriver *drv1;
263 int i;
264
265 drv1 = bdrv_do_find_format(format_name);
266 if (drv1) {
267 return drv1;
268 }
269
270 /* The driver isn't registered, maybe we need to load a module */
271 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
272 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
273 block_module_load_one(block_driver_modules[i].library_name);
274 break;
275 }
276 }
277
278 return bdrv_do_find_format(format_name);
279 }
280
281 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
282 {
283 static const char *whitelist_rw[] = {
284 CONFIG_BDRV_RW_WHITELIST
285 };
286 static const char *whitelist_ro[] = {
287 CONFIG_BDRV_RO_WHITELIST
288 };
289 const char **p;
290
291 if (!whitelist_rw[0] && !whitelist_ro[0]) {
292 return 1; /* no whitelist, anything goes */
293 }
294
295 for (p = whitelist_rw; *p; p++) {
296 if (!strcmp(drv->format_name, *p)) {
297 return 1;
298 }
299 }
300 if (read_only) {
301 for (p = whitelist_ro; *p; p++) {
302 if (!strcmp(drv->format_name, *p)) {
303 return 1;
304 }
305 }
306 }
307 return 0;
308 }
309
310 bool bdrv_uses_whitelist(void)
311 {
312 return use_bdrv_whitelist;
313 }
314
315 typedef struct CreateCo {
316 BlockDriver *drv;
317 char *filename;
318 QemuOpts *opts;
319 int ret;
320 Error *err;
321 } CreateCo;
322
323 static void coroutine_fn bdrv_create_co_entry(void *opaque)
324 {
325 Error *local_err = NULL;
326 int ret;
327
328 CreateCo *cco = opaque;
329 assert(cco->drv);
330
331 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
332 error_propagate(&cco->err, local_err);
333 cco->ret = ret;
334 }
335
336 int bdrv_create(BlockDriver *drv, const char* filename,
337 QemuOpts *opts, Error **errp)
338 {
339 int ret;
340
341 Coroutine *co;
342 CreateCo cco = {
343 .drv = drv,
344 .filename = g_strdup(filename),
345 .opts = opts,
346 .ret = NOT_DONE,
347 .err = NULL,
348 };
349
350 if (!drv->bdrv_create) {
351 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
352 ret = -ENOTSUP;
353 goto out;
354 }
355
356 if (qemu_in_coroutine()) {
357 /* Fast-path if already in coroutine context */
358 bdrv_create_co_entry(&cco);
359 } else {
360 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
361 qemu_coroutine_enter(co);
362 while (cco.ret == NOT_DONE) {
363 aio_poll(qemu_get_aio_context(), true);
364 }
365 }
366
367 ret = cco.ret;
368 if (ret < 0) {
369 if (cco.err) {
370 error_propagate(errp, cco.err);
371 } else {
372 error_setg_errno(errp, -ret, "Could not create image");
373 }
374 }
375
376 out:
377 g_free(cco.filename);
378 return ret;
379 }
380
381 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
382 {
383 BlockDriver *drv;
384 Error *local_err = NULL;
385 int ret;
386
387 drv = bdrv_find_protocol(filename, true, errp);
388 if (drv == NULL) {
389 return -ENOENT;
390 }
391
392 ret = bdrv_create(drv, filename, opts, &local_err);
393 error_propagate(errp, local_err);
394 return ret;
395 }
396
397 /**
398 * Try to get @bs's logical and physical block size.
399 * On success, store them in @bsz struct and return 0.
400 * On failure return -errno.
401 * @bs must not be empty.
402 */
403 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
404 {
405 BlockDriver *drv = bs->drv;
406
407 if (drv && drv->bdrv_probe_blocksizes) {
408 return drv->bdrv_probe_blocksizes(bs, bsz);
409 }
410
411 return -ENOTSUP;
412 }
413
414 /**
415 * Try to get @bs's geometry (cyls, heads, sectors).
416 * On success, store them in @geo struct and return 0.
417 * On failure return -errno.
418 * @bs must not be empty.
419 */
420 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
421 {
422 BlockDriver *drv = bs->drv;
423
424 if (drv && drv->bdrv_probe_geometry) {
425 return drv->bdrv_probe_geometry(bs, geo);
426 }
427
428 return -ENOTSUP;
429 }
430
431 /*
432 * Create a uniquely-named empty temporary file.
433 * Return 0 upon success, otherwise a negative errno value.
434 */
435 int get_tmp_filename(char *filename, int size)
436 {
437 #ifdef _WIN32
438 char temp_dir[MAX_PATH];
439 /* GetTempFileName requires that its output buffer (4th param)
440 have length MAX_PATH or greater. */
441 assert(size >= MAX_PATH);
442 return (GetTempPath(MAX_PATH, temp_dir)
443 && GetTempFileName(temp_dir, "qem", 0, filename)
444 ? 0 : -GetLastError());
445 #else
446 int fd;
447 const char *tmpdir;
448 tmpdir = getenv("TMPDIR");
449 if (!tmpdir) {
450 tmpdir = "/var/tmp";
451 }
452 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
453 return -EOVERFLOW;
454 }
455 fd = mkstemp(filename);
456 if (fd < 0) {
457 return -errno;
458 }
459 if (close(fd) != 0) {
460 unlink(filename);
461 return -errno;
462 }
463 return 0;
464 #endif
465 }
466
467 /*
468 * Detect host devices. By convention, /dev/cdrom[N] is always
469 * recognized as a host CDROM.
470 */
471 static BlockDriver *find_hdev_driver(const char *filename)
472 {
473 int score_max = 0, score;
474 BlockDriver *drv = NULL, *d;
475
476 QLIST_FOREACH(d, &bdrv_drivers, list) {
477 if (d->bdrv_probe_device) {
478 score = d->bdrv_probe_device(filename);
479 if (score > score_max) {
480 score_max = score;
481 drv = d;
482 }
483 }
484 }
485
486 return drv;
487 }
488
489 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
490 {
491 BlockDriver *drv1;
492
493 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
494 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
495 return drv1;
496 }
497 }
498
499 return NULL;
500 }
501
502 BlockDriver *bdrv_find_protocol(const char *filename,
503 bool allow_protocol_prefix,
504 Error **errp)
505 {
506 BlockDriver *drv1;
507 char protocol[128];
508 int len;
509 const char *p;
510 int i;
511
512 /* TODO Drivers without bdrv_file_open must be specified explicitly */
513
514 /*
515 * XXX(hch): we really should not let host device detection
516 * override an explicit protocol specification, but moving this
517 * later breaks access to device names with colons in them.
518 * Thanks to the brain-dead persistent naming schemes on udev-
519 * based Linux systems those actually are quite common.
520 */
521 drv1 = find_hdev_driver(filename);
522 if (drv1) {
523 return drv1;
524 }
525
526 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
527 return &bdrv_file;
528 }
529
530 p = strchr(filename, ':');
531 assert(p != NULL);
532 len = p - filename;
533 if (len > sizeof(protocol) - 1)
534 len = sizeof(protocol) - 1;
535 memcpy(protocol, filename, len);
536 protocol[len] = '\0';
537
538 drv1 = bdrv_do_find_protocol(protocol);
539 if (drv1) {
540 return drv1;
541 }
542
543 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
544 if (block_driver_modules[i].protocol_name &&
545 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
546 block_module_load_one(block_driver_modules[i].library_name);
547 break;
548 }
549 }
550
551 drv1 = bdrv_do_find_protocol(protocol);
552 if (!drv1) {
553 error_setg(errp, "Unknown protocol '%s'", protocol);
554 }
555 return drv1;
556 }
557
558 /*
559 * Guess image format by probing its contents.
560 * This is not a good idea when your image is raw (CVE-2008-2004), but
561 * we do it anyway for backward compatibility.
562 *
563 * @buf contains the image's first @buf_size bytes.
564 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
565 * but can be smaller if the image file is smaller)
566 * @filename is its filename.
567 *
568 * For all block drivers, call the bdrv_probe() method to get its
569 * probing score.
570 * Return the first block driver with the highest probing score.
571 */
572 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
573 const char *filename)
574 {
575 int score_max = 0, score;
576 BlockDriver *drv = NULL, *d;
577
578 QLIST_FOREACH(d, &bdrv_drivers, list) {
579 if (d->bdrv_probe) {
580 score = d->bdrv_probe(buf, buf_size, filename);
581 if (score > score_max) {
582 score_max = score;
583 drv = d;
584 }
585 }
586 }
587
588 return drv;
589 }
590
591 static int find_image_format(BlockBackend *file, const char *filename,
592 BlockDriver **pdrv, Error **errp)
593 {
594 BlockDriver *drv;
595 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
596 int ret = 0;
597
598 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
599 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
600 *pdrv = &bdrv_raw;
601 return ret;
602 }
603
604 ret = blk_pread(file, 0, buf, sizeof(buf));
605 if (ret < 0) {
606 error_setg_errno(errp, -ret, "Could not read image for determining its "
607 "format");
608 *pdrv = NULL;
609 return ret;
610 }
611
612 drv = bdrv_probe_all(buf, ret, filename);
613 if (!drv) {
614 error_setg(errp, "Could not determine image format: No compatible "
615 "driver found");
616 ret = -ENOENT;
617 }
618 *pdrv = drv;
619 return ret;
620 }
621
622 /**
623 * Set the current 'total_sectors' value
624 * Return 0 on success, -errno on error.
625 */
626 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
627 {
628 BlockDriver *drv = bs->drv;
629
630 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
631 if (bdrv_is_sg(bs))
632 return 0;
633
634 /* query actual device if possible, otherwise just trust the hint */
635 if (drv->bdrv_getlength) {
636 int64_t length = drv->bdrv_getlength(bs);
637 if (length < 0) {
638 return length;
639 }
640 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
641 }
642
643 bs->total_sectors = hint;
644 return 0;
645 }
646
647 /**
648 * Combines a QDict of new block driver @options with any missing options taken
649 * from @old_options, so that leaving out an option defaults to its old value.
650 */
651 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
652 QDict *old_options)
653 {
654 if (bs->drv && bs->drv->bdrv_join_options) {
655 bs->drv->bdrv_join_options(options, old_options);
656 } else {
657 qdict_join(options, old_options, false);
658 }
659 }
660
661 /**
662 * Set open flags for a given discard mode
663 *
664 * Return 0 on success, -1 if the discard mode was invalid.
665 */
666 int bdrv_parse_discard_flags(const char *mode, int *flags)
667 {
668 *flags &= ~BDRV_O_UNMAP;
669
670 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
671 /* do nothing */
672 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
673 *flags |= BDRV_O_UNMAP;
674 } else {
675 return -1;
676 }
677
678 return 0;
679 }
680
681 /**
682 * Set open flags for a given cache mode
683 *
684 * Return 0 on success, -1 if the cache mode was invalid.
685 */
686 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
687 {
688 *flags &= ~BDRV_O_CACHE_MASK;
689
690 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
691 *writethrough = false;
692 *flags |= BDRV_O_NOCACHE;
693 } else if (!strcmp(mode, "directsync")) {
694 *writethrough = true;
695 *flags |= BDRV_O_NOCACHE;
696 } else if (!strcmp(mode, "writeback")) {
697 *writethrough = false;
698 } else if (!strcmp(mode, "unsafe")) {
699 *writethrough = false;
700 *flags |= BDRV_O_NO_FLUSH;
701 } else if (!strcmp(mode, "writethrough")) {
702 *writethrough = true;
703 } else {
704 return -1;
705 }
706
707 return 0;
708 }
709
710 static char *bdrv_child_get_parent_desc(BdrvChild *c)
711 {
712 BlockDriverState *parent = c->opaque;
713 return g_strdup(bdrv_get_device_or_node_name(parent));
714 }
715
716 static void bdrv_child_cb_drained_begin(BdrvChild *child)
717 {
718 BlockDriverState *bs = child->opaque;
719 bdrv_drained_begin(bs);
720 }
721
722 static void bdrv_child_cb_drained_end(BdrvChild *child)
723 {
724 BlockDriverState *bs = child->opaque;
725 bdrv_drained_end(bs);
726 }
727
728 /*
729 * Returns the options and flags that a temporary snapshot should get, based on
730 * the originally requested flags (the originally requested image will have
731 * flags like a backing file)
732 */
733 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
734 int parent_flags, QDict *parent_options)
735 {
736 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
737
738 /* For temporary files, unconditional cache=unsafe is fine */
739 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
740 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
741
742 /* Copy the read-only option from the parent */
743 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
744
745 /* aio=native doesn't work for cache.direct=off, so disable it for the
746 * temporary snapshot */
747 *child_flags &= ~BDRV_O_NATIVE_AIO;
748 }
749
750 /*
751 * Returns the options and flags that bs->file should get if a protocol driver
752 * is expected, based on the given options and flags for the parent BDS
753 */
754 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
755 int parent_flags, QDict *parent_options)
756 {
757 int flags = parent_flags;
758
759 /* Enable protocol handling, disable format probing for bs->file */
760 flags |= BDRV_O_PROTOCOL;
761
762 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
763 * the parent. */
764 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
765 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
766
767 /* Inherit the read-only option from the parent if it's not set */
768 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
769
770 /* Our block drivers take care to send flushes and respect unmap policy,
771 * so we can default to enable both on lower layers regardless of the
772 * corresponding parent options. */
773 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
774
775 /* Clear flags that only apply to the top layer */
776 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
777 BDRV_O_NO_IO);
778
779 *child_flags = flags;
780 }
781
782 const BdrvChildRole child_file = {
783 .get_parent_desc = bdrv_child_get_parent_desc,
784 .inherit_options = bdrv_inherited_options,
785 .drained_begin = bdrv_child_cb_drained_begin,
786 .drained_end = bdrv_child_cb_drained_end,
787 };
788
789 /*
790 * Returns the options and flags that bs->file should get if the use of formats
791 * (and not only protocols) is permitted for it, based on the given options and
792 * flags for the parent BDS
793 */
794 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
795 int parent_flags, QDict *parent_options)
796 {
797 child_file.inherit_options(child_flags, child_options,
798 parent_flags, parent_options);
799
800 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
801 }
802
803 const BdrvChildRole child_format = {
804 .get_parent_desc = bdrv_child_get_parent_desc,
805 .inherit_options = bdrv_inherited_fmt_options,
806 .drained_begin = bdrv_child_cb_drained_begin,
807 .drained_end = bdrv_child_cb_drained_end,
808 };
809
810 static void bdrv_backing_attach(BdrvChild *c)
811 {
812 BlockDriverState *parent = c->opaque;
813 BlockDriverState *backing_hd = c->bs;
814
815 assert(!parent->backing_blocker);
816 error_setg(&parent->backing_blocker,
817 "node is used as backing hd of '%s'",
818 bdrv_get_device_or_node_name(parent));
819
820 parent->open_flags &= ~BDRV_O_NO_BACKING;
821 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
822 backing_hd->filename);
823 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
824 backing_hd->drv ? backing_hd->drv->format_name : "");
825
826 bdrv_op_block_all(backing_hd, parent->backing_blocker);
827 /* Otherwise we won't be able to commit or stream */
828 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
829 parent->backing_blocker);
830 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
831 parent->backing_blocker);
832 /*
833 * We do backup in 3 ways:
834 * 1. drive backup
835 * The target bs is new opened, and the source is top BDS
836 * 2. blockdev backup
837 * Both the source and the target are top BDSes.
838 * 3. internal backup(used for block replication)
839 * Both the source and the target are backing file
840 *
841 * In case 1 and 2, neither the source nor the target is the backing file.
842 * In case 3, we will block the top BDS, so there is only one block job
843 * for the top BDS and its backing chain.
844 */
845 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
846 parent->backing_blocker);
847 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
848 parent->backing_blocker);
849 }
850
851 static void bdrv_backing_detach(BdrvChild *c)
852 {
853 BlockDriverState *parent = c->opaque;
854
855 assert(parent->backing_blocker);
856 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
857 error_free(parent->backing_blocker);
858 parent->backing_blocker = NULL;
859 }
860
861 /*
862 * Returns the options and flags that bs->backing should get, based on the
863 * given options and flags for the parent BDS
864 */
865 static void bdrv_backing_options(int *child_flags, QDict *child_options,
866 int parent_flags, QDict *parent_options)
867 {
868 int flags = parent_flags;
869
870 /* The cache mode is inherited unmodified for backing files; except WCE,
871 * which is only applied on the top level (BlockBackend) */
872 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
873 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
874
875 /* backing files always opened read-only */
876 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
877 flags &= ~BDRV_O_COPY_ON_READ;
878
879 /* snapshot=on is handled on the top layer */
880 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
881
882 *child_flags = flags;
883 }
884
885 const BdrvChildRole child_backing = {
886 .get_parent_desc = bdrv_child_get_parent_desc,
887 .attach = bdrv_backing_attach,
888 .detach = bdrv_backing_detach,
889 .inherit_options = bdrv_backing_options,
890 .drained_begin = bdrv_child_cb_drained_begin,
891 .drained_end = bdrv_child_cb_drained_end,
892 };
893
894 static int bdrv_open_flags(BlockDriverState *bs, int flags)
895 {
896 int open_flags = flags;
897
898 /*
899 * Clear flags that are internal to the block layer before opening the
900 * image.
901 */
902 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
903
904 /*
905 * Snapshots should be writable.
906 */
907 if (flags & BDRV_O_TEMPORARY) {
908 open_flags |= BDRV_O_RDWR;
909 }
910
911 return open_flags;
912 }
913
914 static void update_flags_from_options(int *flags, QemuOpts *opts)
915 {
916 *flags &= ~BDRV_O_CACHE_MASK;
917
918 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
919 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
920 *flags |= BDRV_O_NO_FLUSH;
921 }
922
923 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
924 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
925 *flags |= BDRV_O_NOCACHE;
926 }
927
928 *flags &= ~BDRV_O_RDWR;
929
930 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
931 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
932 *flags |= BDRV_O_RDWR;
933 }
934
935 }
936
937 static void update_options_from_flags(QDict *options, int flags)
938 {
939 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
940 qdict_put(options, BDRV_OPT_CACHE_DIRECT,
941 qbool_from_bool(flags & BDRV_O_NOCACHE));
942 }
943 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
944 qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
945 qbool_from_bool(flags & BDRV_O_NO_FLUSH));
946 }
947 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
948 qdict_put(options, BDRV_OPT_READ_ONLY,
949 qbool_from_bool(!(flags & BDRV_O_RDWR)));
950 }
951 }
952
953 static void bdrv_assign_node_name(BlockDriverState *bs,
954 const char *node_name,
955 Error **errp)
956 {
957 char *gen_node_name = NULL;
958
959 if (!node_name) {
960 node_name = gen_node_name = id_generate(ID_BLOCK);
961 } else if (!id_wellformed(node_name)) {
962 /*
963 * Check for empty string or invalid characters, but not if it is
964 * generated (generated names use characters not available to the user)
965 */
966 error_setg(errp, "Invalid node name");
967 return;
968 }
969
970 /* takes care of avoiding namespaces collisions */
971 if (blk_by_name(node_name)) {
972 error_setg(errp, "node-name=%s is conflicting with a device id",
973 node_name);
974 goto out;
975 }
976
977 /* takes care of avoiding duplicates node names */
978 if (bdrv_find_node(node_name)) {
979 error_setg(errp, "Duplicate node name");
980 goto out;
981 }
982
983 /* copy node name into the bs and insert it into the graph list */
984 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
985 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
986 out:
987 g_free(gen_node_name);
988 }
989
990 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
991 const char *node_name, QDict *options,
992 int open_flags, Error **errp)
993 {
994 Error *local_err = NULL;
995 int ret;
996
997 bdrv_assign_node_name(bs, node_name, &local_err);
998 if (local_err) {
999 error_propagate(errp, local_err);
1000 return -EINVAL;
1001 }
1002
1003 bs->drv = drv;
1004 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1005 bs->opaque = g_malloc0(drv->instance_size);
1006
1007 if (drv->bdrv_file_open) {
1008 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1009 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1010 } else if (drv->bdrv_open) {
1011 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1012 } else {
1013 ret = 0;
1014 }
1015
1016 if (ret < 0) {
1017 if (local_err) {
1018 error_propagate(errp, local_err);
1019 } else if (bs->filename[0]) {
1020 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1021 } else {
1022 error_setg_errno(errp, -ret, "Could not open image");
1023 }
1024 goto free_and_fail;
1025 }
1026
1027 ret = refresh_total_sectors(bs, bs->total_sectors);
1028 if (ret < 0) {
1029 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1030 goto free_and_fail;
1031 }
1032
1033 bdrv_refresh_limits(bs, &local_err);
1034 if (local_err) {
1035 error_propagate(errp, local_err);
1036 ret = -EINVAL;
1037 goto free_and_fail;
1038 }
1039
1040 assert(bdrv_opt_mem_align(bs) != 0);
1041 assert(bdrv_min_mem_align(bs) != 0);
1042 assert(is_power_of_2(bs->bl.request_alignment));
1043
1044 return 0;
1045
1046 free_and_fail:
1047 /* FIXME Close bs first if already opened*/
1048 g_free(bs->opaque);
1049 bs->opaque = NULL;
1050 bs->drv = NULL;
1051 return ret;
1052 }
1053
1054 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1055 int flags, Error **errp)
1056 {
1057 BlockDriverState *bs;
1058 int ret;
1059
1060 bs = bdrv_new();
1061 bs->open_flags = flags;
1062 bs->explicit_options = qdict_new();
1063 bs->options = qdict_new();
1064 bs->opaque = NULL;
1065
1066 update_options_from_flags(bs->options, flags);
1067
1068 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1069 if (ret < 0) {
1070 QDECREF(bs->explicit_options);
1071 QDECREF(bs->options);
1072 bdrv_unref(bs);
1073 return NULL;
1074 }
1075
1076 return bs;
1077 }
1078
1079 QemuOptsList bdrv_runtime_opts = {
1080 .name = "bdrv_common",
1081 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1082 .desc = {
1083 {
1084 .name = "node-name",
1085 .type = QEMU_OPT_STRING,
1086 .help = "Node name of the block device node",
1087 },
1088 {
1089 .name = "driver",
1090 .type = QEMU_OPT_STRING,
1091 .help = "Block driver to use for the node",
1092 },
1093 {
1094 .name = BDRV_OPT_CACHE_DIRECT,
1095 .type = QEMU_OPT_BOOL,
1096 .help = "Bypass software writeback cache on the host",
1097 },
1098 {
1099 .name = BDRV_OPT_CACHE_NO_FLUSH,
1100 .type = QEMU_OPT_BOOL,
1101 .help = "Ignore flush requests",
1102 },
1103 {
1104 .name = BDRV_OPT_READ_ONLY,
1105 .type = QEMU_OPT_BOOL,
1106 .help = "Node is opened in read-only mode",
1107 },
1108 {
1109 .name = "detect-zeroes",
1110 .type = QEMU_OPT_STRING,
1111 .help = "try to optimize zero writes (off, on, unmap)",
1112 },
1113 {
1114 .name = "discard",
1115 .type = QEMU_OPT_STRING,
1116 .help = "discard operation (ignore/off, unmap/on)",
1117 },
1118 { /* end of list */ }
1119 },
1120 };
1121
1122 /*
1123 * Common part for opening disk images and files
1124 *
1125 * Removes all processed options from *options.
1126 */
1127 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1128 QDict *options, Error **errp)
1129 {
1130 int ret, open_flags;
1131 const char *filename;
1132 const char *driver_name = NULL;
1133 const char *node_name = NULL;
1134 const char *discard;
1135 const char *detect_zeroes;
1136 QemuOpts *opts;
1137 BlockDriver *drv;
1138 Error *local_err = NULL;
1139
1140 assert(bs->file == NULL);
1141 assert(options != NULL && bs->options != options);
1142
1143 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1144 qemu_opts_absorb_qdict(opts, options, &local_err);
1145 if (local_err) {
1146 error_propagate(errp, local_err);
1147 ret = -EINVAL;
1148 goto fail_opts;
1149 }
1150
1151 update_flags_from_options(&bs->open_flags, opts);
1152
1153 driver_name = qemu_opt_get(opts, "driver");
1154 drv = bdrv_find_format(driver_name);
1155 assert(drv != NULL);
1156
1157 if (file != NULL) {
1158 filename = blk_bs(file)->filename;
1159 } else {
1160 filename = qdict_get_try_str(options, "filename");
1161 }
1162
1163 if (drv->bdrv_needs_filename && !filename) {
1164 error_setg(errp, "The '%s' block driver requires a file name",
1165 drv->format_name);
1166 ret = -EINVAL;
1167 goto fail_opts;
1168 }
1169
1170 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1171 drv->format_name);
1172
1173 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1174
1175 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1176 error_setg(errp,
1177 !bs->read_only && bdrv_is_whitelisted(drv, true)
1178 ? "Driver '%s' can only be used for read-only devices"
1179 : "Driver '%s' is not whitelisted",
1180 drv->format_name);
1181 ret = -ENOTSUP;
1182 goto fail_opts;
1183 }
1184
1185 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1186 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1187 if (!bs->read_only) {
1188 bdrv_enable_copy_on_read(bs);
1189 } else {
1190 error_setg(errp, "Can't use copy-on-read on read-only device");
1191 ret = -EINVAL;
1192 goto fail_opts;
1193 }
1194 }
1195
1196 discard = qemu_opt_get(opts, "discard");
1197 if (discard != NULL) {
1198 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1199 error_setg(errp, "Invalid discard option");
1200 ret = -EINVAL;
1201 goto fail_opts;
1202 }
1203 }
1204
1205 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1206 if (detect_zeroes) {
1207 BlockdevDetectZeroesOptions value =
1208 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1209 detect_zeroes,
1210 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1211 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1212 &local_err);
1213 if (local_err) {
1214 error_propagate(errp, local_err);
1215 ret = -EINVAL;
1216 goto fail_opts;
1217 }
1218
1219 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1220 !(bs->open_flags & BDRV_O_UNMAP))
1221 {
1222 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1223 "without setting discard operation to unmap");
1224 ret = -EINVAL;
1225 goto fail_opts;
1226 }
1227
1228 bs->detect_zeroes = value;
1229 }
1230
1231 if (filename != NULL) {
1232 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1233 } else {
1234 bs->filename[0] = '\0';
1235 }
1236 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1237
1238 /* Open the image, either directly or using a protocol */
1239 open_flags = bdrv_open_flags(bs, bs->open_flags);
1240 node_name = qemu_opt_get(opts, "node-name");
1241
1242 assert(!drv->bdrv_file_open || file == NULL);
1243 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1244 if (ret < 0) {
1245 goto fail_opts;
1246 }
1247
1248 qemu_opts_del(opts);
1249 return 0;
1250
1251 fail_opts:
1252 qemu_opts_del(opts);
1253 return ret;
1254 }
1255
1256 static QDict *parse_json_filename(const char *filename, Error **errp)
1257 {
1258 QObject *options_obj;
1259 QDict *options;
1260 int ret;
1261
1262 ret = strstart(filename, "json:", &filename);
1263 assert(ret);
1264
1265 options_obj = qobject_from_json(filename, errp);
1266 if (!options_obj) {
1267 /* Work around qobject_from_json() lossage TODO fix that */
1268 if (errp && !*errp) {
1269 error_setg(errp, "Could not parse the JSON options");
1270 return NULL;
1271 }
1272 error_prepend(errp, "Could not parse the JSON options: ");
1273 return NULL;
1274 }
1275
1276 options = qobject_to_qdict(options_obj);
1277 if (!options) {
1278 qobject_decref(options_obj);
1279 error_setg(errp, "Invalid JSON object given");
1280 return NULL;
1281 }
1282
1283 qdict_flatten(options);
1284
1285 return options;
1286 }
1287
1288 static void parse_json_protocol(QDict *options, const char **pfilename,
1289 Error **errp)
1290 {
1291 QDict *json_options;
1292 Error *local_err = NULL;
1293
1294 /* Parse json: pseudo-protocol */
1295 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1296 return;
1297 }
1298
1299 json_options = parse_json_filename(*pfilename, &local_err);
1300 if (local_err) {
1301 error_propagate(errp, local_err);
1302 return;
1303 }
1304
1305 /* Options given in the filename have lower priority than options
1306 * specified directly */
1307 qdict_join(options, json_options, false);
1308 QDECREF(json_options);
1309 *pfilename = NULL;
1310 }
1311
1312 /*
1313 * Fills in default options for opening images and converts the legacy
1314 * filename/flags pair to option QDict entries.
1315 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1316 * block driver has been specified explicitly.
1317 */
1318 static int bdrv_fill_options(QDict **options, const char *filename,
1319 int *flags, Error **errp)
1320 {
1321 const char *drvname;
1322 bool protocol = *flags & BDRV_O_PROTOCOL;
1323 bool parse_filename = false;
1324 BlockDriver *drv = NULL;
1325 Error *local_err = NULL;
1326
1327 drvname = qdict_get_try_str(*options, "driver");
1328 if (drvname) {
1329 drv = bdrv_find_format(drvname);
1330 if (!drv) {
1331 error_setg(errp, "Unknown driver '%s'", drvname);
1332 return -ENOENT;
1333 }
1334 /* If the user has explicitly specified the driver, this choice should
1335 * override the BDRV_O_PROTOCOL flag */
1336 protocol = drv->bdrv_file_open;
1337 }
1338
1339 if (protocol) {
1340 *flags |= BDRV_O_PROTOCOL;
1341 } else {
1342 *flags &= ~BDRV_O_PROTOCOL;
1343 }
1344
1345 /* Translate cache options from flags into options */
1346 update_options_from_flags(*options, *flags);
1347
1348 /* Fetch the file name from the options QDict if necessary */
1349 if (protocol && filename) {
1350 if (!qdict_haskey(*options, "filename")) {
1351 qdict_put(*options, "filename", qstring_from_str(filename));
1352 parse_filename = true;
1353 } else {
1354 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1355 "the same time");
1356 return -EINVAL;
1357 }
1358 }
1359
1360 /* Find the right block driver */
1361 filename = qdict_get_try_str(*options, "filename");
1362
1363 if (!drvname && protocol) {
1364 if (filename) {
1365 drv = bdrv_find_protocol(filename, parse_filename, errp);
1366 if (!drv) {
1367 return -EINVAL;
1368 }
1369
1370 drvname = drv->format_name;
1371 qdict_put(*options, "driver", qstring_from_str(drvname));
1372 } else {
1373 error_setg(errp, "Must specify either driver or file");
1374 return -EINVAL;
1375 }
1376 }
1377
1378 assert(drv || !protocol);
1379
1380 /* Driver-specific filename parsing */
1381 if (drv && drv->bdrv_parse_filename && parse_filename) {
1382 drv->bdrv_parse_filename(filename, *options, &local_err);
1383 if (local_err) {
1384 error_propagate(errp, local_err);
1385 return -EINVAL;
1386 }
1387
1388 if (!drv->bdrv_needs_filename) {
1389 qdict_del(*options, "filename");
1390 }
1391 }
1392
1393 return 0;
1394 }
1395
1396 /*
1397 * Check whether permissions on this node can be changed in a way that
1398 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1399 * permissions of all its parents. This involves checking whether all necessary
1400 * permission changes to child nodes can be performed.
1401 *
1402 * A call to this function must always be followed by a call to bdrv_set_perm()
1403 * or bdrv_abort_perm_update().
1404 */
1405 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1406 uint64_t cumulative_shared_perms,
1407 GSList *ignore_children, Error **errp)
1408 {
1409 BlockDriver *drv = bs->drv;
1410 BdrvChild *c;
1411 int ret;
1412
1413 /* Write permissions never work with read-only images */
1414 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1415 bdrv_is_read_only(bs))
1416 {
1417 error_setg(errp, "Block node is read-only");
1418 return -EPERM;
1419 }
1420
1421 /* Check this node */
1422 if (!drv) {
1423 return 0;
1424 }
1425
1426 if (drv->bdrv_check_perm) {
1427 return drv->bdrv_check_perm(bs, cumulative_perms,
1428 cumulative_shared_perms, errp);
1429 }
1430
1431 /* Drivers that never have children can omit .bdrv_child_perm() */
1432 if (!drv->bdrv_child_perm) {
1433 assert(QLIST_EMPTY(&bs->children));
1434 return 0;
1435 }
1436
1437 /* Check all children */
1438 QLIST_FOREACH(c, &bs->children, next) {
1439 uint64_t cur_perm, cur_shared;
1440 drv->bdrv_child_perm(bs, c, c->role,
1441 cumulative_perms, cumulative_shared_perms,
1442 &cur_perm, &cur_shared);
1443 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1444 errp);
1445 if (ret < 0) {
1446 return ret;
1447 }
1448 }
1449
1450 return 0;
1451 }
1452
1453 /*
1454 * Notifies drivers that after a previous bdrv_check_perm() call, the
1455 * permission update is not performed and any preparations made for it (e.g.
1456 * taken file locks) need to be undone.
1457 *
1458 * This function recursively notifies all child nodes.
1459 */
1460 static void bdrv_abort_perm_update(BlockDriverState *bs)
1461 {
1462 BlockDriver *drv = bs->drv;
1463 BdrvChild *c;
1464
1465 if (!drv) {
1466 return;
1467 }
1468
1469 if (drv->bdrv_abort_perm_update) {
1470 drv->bdrv_abort_perm_update(bs);
1471 }
1472
1473 QLIST_FOREACH(c, &bs->children, next) {
1474 bdrv_child_abort_perm_update(c);
1475 }
1476 }
1477
1478 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1479 uint64_t cumulative_shared_perms)
1480 {
1481 BlockDriver *drv = bs->drv;
1482 BdrvChild *c;
1483
1484 if (!drv) {
1485 return;
1486 }
1487
1488 /* Update this node */
1489 if (drv->bdrv_set_perm) {
1490 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1491 }
1492
1493 /* Drivers that never have children can omit .bdrv_child_perm() */
1494 if (!drv->bdrv_child_perm) {
1495 assert(QLIST_EMPTY(&bs->children));
1496 return;
1497 }
1498
1499 /* Update all children */
1500 QLIST_FOREACH(c, &bs->children, next) {
1501 uint64_t cur_perm, cur_shared;
1502 drv->bdrv_child_perm(bs, c, c->role,
1503 cumulative_perms, cumulative_shared_perms,
1504 &cur_perm, &cur_shared);
1505 bdrv_child_set_perm(c, cur_perm, cur_shared);
1506 }
1507 }
1508
1509 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1510 uint64_t *shared_perm)
1511 {
1512 BdrvChild *c;
1513 uint64_t cumulative_perms = 0;
1514 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1515
1516 QLIST_FOREACH(c, &bs->parents, next_parent) {
1517 cumulative_perms |= c->perm;
1518 cumulative_shared_perms &= c->shared_perm;
1519 }
1520
1521 *perm = cumulative_perms;
1522 *shared_perm = cumulative_shared_perms;
1523 }
1524
1525 static char *bdrv_child_user_desc(BdrvChild *c)
1526 {
1527 if (c->role->get_parent_desc) {
1528 return c->role->get_parent_desc(c);
1529 }
1530
1531 return g_strdup("another user");
1532 }
1533
1534 static char *bdrv_perm_names(uint64_t perm)
1535 {
1536 struct perm_name {
1537 uint64_t perm;
1538 const char *name;
1539 } permissions[] = {
1540 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1541 { BLK_PERM_WRITE, "write" },
1542 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1543 { BLK_PERM_RESIZE, "resize" },
1544 { BLK_PERM_GRAPH_MOD, "change children" },
1545 { 0, NULL }
1546 };
1547
1548 char *result = g_strdup("");
1549 struct perm_name *p;
1550
1551 for (p = permissions; p->name; p++) {
1552 if (perm & p->perm) {
1553 char *old = result;
1554 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1555 g_free(old);
1556 }
1557 }
1558
1559 return result;
1560 }
1561
1562 /*
1563 * Checks whether a new reference to @bs can be added if the new user requires
1564 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1565 * set, the BdrvChild objects in this list are ignored in the calculations;
1566 * this allows checking permission updates for an existing reference.
1567 *
1568 * Needs to be followed by a call to either bdrv_set_perm() or
1569 * bdrv_abort_perm_update(). */
1570 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1571 uint64_t new_shared_perm,
1572 GSList *ignore_children, Error **errp)
1573 {
1574 BdrvChild *c;
1575 uint64_t cumulative_perms = new_used_perm;
1576 uint64_t cumulative_shared_perms = new_shared_perm;
1577
1578 /* There is no reason why anyone couldn't tolerate write_unchanged */
1579 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1580
1581 QLIST_FOREACH(c, &bs->parents, next_parent) {
1582 if (g_slist_find(ignore_children, c)) {
1583 continue;
1584 }
1585
1586 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1587 char *user = bdrv_child_user_desc(c);
1588 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1589 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1590 "allow '%s' on %s",
1591 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1592 g_free(user);
1593 g_free(perm_names);
1594 return -EPERM;
1595 }
1596
1597 if ((c->perm & new_shared_perm) != c->perm) {
1598 char *user = bdrv_child_user_desc(c);
1599 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1600 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1601 "'%s' on %s",
1602 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1603 g_free(user);
1604 g_free(perm_names);
1605 return -EPERM;
1606 }
1607
1608 cumulative_perms |= c->perm;
1609 cumulative_shared_perms &= c->shared_perm;
1610 }
1611
1612 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1613 ignore_children, errp);
1614 }
1615
1616 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1617 * bdrv_child_abort_perm_update(). */
1618 int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1619 GSList *ignore_children, Error **errp)
1620 {
1621 int ret;
1622
1623 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1624 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1625 g_slist_free(ignore_children);
1626
1627 return ret;
1628 }
1629
1630 void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1631 {
1632 uint64_t cumulative_perms, cumulative_shared_perms;
1633
1634 c->perm = perm;
1635 c->shared_perm = shared;
1636
1637 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1638 &cumulative_shared_perms);
1639 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1640 }
1641
1642 void bdrv_child_abort_perm_update(BdrvChild *c)
1643 {
1644 bdrv_abort_perm_update(c->bs);
1645 }
1646
1647 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1648 Error **errp)
1649 {
1650 int ret;
1651
1652 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1653 if (ret < 0) {
1654 bdrv_child_abort_perm_update(c);
1655 return ret;
1656 }
1657
1658 bdrv_child_set_perm(c, perm, shared);
1659
1660 return 0;
1661 }
1662
1663 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1664 | BLK_PERM_WRITE \
1665 | BLK_PERM_WRITE_UNCHANGED \
1666 | BLK_PERM_RESIZE)
1667 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1668
1669 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1670 const BdrvChildRole *role,
1671 uint64_t perm, uint64_t shared,
1672 uint64_t *nperm, uint64_t *nshared)
1673 {
1674 if (c == NULL) {
1675 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1676 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1677 return;
1678 }
1679
1680 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1681 (c->perm & DEFAULT_PERM_UNCHANGED);
1682 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1683 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1684 }
1685
1686 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1687 const BdrvChildRole *role,
1688 uint64_t perm, uint64_t shared,
1689 uint64_t *nperm, uint64_t *nshared)
1690 {
1691 bool backing = (role == &child_backing);
1692 assert(role == &child_backing || role == &child_file);
1693
1694 if (!backing) {
1695 /* Apart from the modifications below, the same permissions are
1696 * forwarded and left alone as for filters */
1697 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1698
1699 /* Format drivers may touch metadata even if the guest doesn't write */
1700 if (!bdrv_is_read_only(bs)) {
1701 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1702 }
1703
1704 /* bs->file always needs to be consistent because of the metadata. We
1705 * can never allow other users to resize or write to it. */
1706 perm |= BLK_PERM_CONSISTENT_READ;
1707 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1708 } else {
1709 /* We want consistent read from backing files if the parent needs it.
1710 * No other operations are performed on backing files. */
1711 perm &= BLK_PERM_CONSISTENT_READ;
1712
1713 /* If the parent can deal with changing data, we're okay with a
1714 * writable and resizable backing file. */
1715 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1716 if (shared & BLK_PERM_WRITE) {
1717 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1718 } else {
1719 shared = 0;
1720 }
1721
1722 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1723 BLK_PERM_WRITE_UNCHANGED;
1724 }
1725
1726 *nperm = perm;
1727 *nshared = shared;
1728 }
1729
1730 static void bdrv_replace_child_noperm(BdrvChild *child,
1731 BlockDriverState *new_bs)
1732 {
1733 BlockDriverState *old_bs = child->bs;
1734
1735 if (old_bs) {
1736 if (old_bs->quiesce_counter && child->role->drained_end) {
1737 child->role->drained_end(child);
1738 }
1739 if (child->role->detach) {
1740 child->role->detach(child);
1741 }
1742 QLIST_REMOVE(child, next_parent);
1743 }
1744
1745 child->bs = new_bs;
1746
1747 if (new_bs) {
1748 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1749 if (new_bs->quiesce_counter && child->role->drained_begin) {
1750 child->role->drained_begin(child);
1751 }
1752
1753 if (child->role->attach) {
1754 child->role->attach(child);
1755 }
1756 }
1757 }
1758
1759 /*
1760 * Updates @child to change its reference to point to @new_bs, including
1761 * checking and applying the necessary permisson updates both to the old node
1762 * and to @new_bs.
1763 *
1764 * NULL is passed as @new_bs for removing the reference before freeing @child.
1765 *
1766 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1767 * function uses bdrv_set_perm() to update the permissions according to the new
1768 * reference that @new_bs gets.
1769 */
1770 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1771 {
1772 BlockDriverState *old_bs = child->bs;
1773 uint64_t perm, shared_perm;
1774
1775 if (old_bs) {
1776 /* Update permissions for old node. This is guaranteed to succeed
1777 * because we're just taking a parent away, so we're loosening
1778 * restrictions. */
1779 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1780 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1781 bdrv_set_perm(old_bs, perm, shared_perm);
1782 }
1783
1784 bdrv_replace_child_noperm(child, new_bs);
1785
1786 if (new_bs) {
1787 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1788 bdrv_set_perm(new_bs, perm, shared_perm);
1789 }
1790 }
1791
1792 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1793 const char *child_name,
1794 const BdrvChildRole *child_role,
1795 uint64_t perm, uint64_t shared_perm,
1796 void *opaque, Error **errp)
1797 {
1798 BdrvChild *child;
1799 int ret;
1800
1801 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1802 if (ret < 0) {
1803 bdrv_abort_perm_update(child_bs);
1804 return NULL;
1805 }
1806
1807 child = g_new(BdrvChild, 1);
1808 *child = (BdrvChild) {
1809 .bs = NULL,
1810 .name = g_strdup(child_name),
1811 .role = child_role,
1812 .perm = perm,
1813 .shared_perm = shared_perm,
1814 .opaque = opaque,
1815 };
1816
1817 /* This performs the matching bdrv_set_perm() for the above check. */
1818 bdrv_replace_child(child, child_bs);
1819
1820 return child;
1821 }
1822
1823 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1824 BlockDriverState *child_bs,
1825 const char *child_name,
1826 const BdrvChildRole *child_role,
1827 Error **errp)
1828 {
1829 BdrvChild *child;
1830 uint64_t perm, shared_perm;
1831
1832 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1833
1834 assert(parent_bs->drv);
1835 parent_bs->drv->bdrv_child_perm(parent_bs, NULL, child_role,
1836 perm, shared_perm, &perm, &shared_perm);
1837
1838 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1839 perm, shared_perm, parent_bs, errp);
1840 if (child == NULL) {
1841 return NULL;
1842 }
1843
1844 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1845 return child;
1846 }
1847
1848 static void bdrv_detach_child(BdrvChild *child)
1849 {
1850 if (child->next.le_prev) {
1851 QLIST_REMOVE(child, next);
1852 child->next.le_prev = NULL;
1853 }
1854
1855 bdrv_replace_child(child, NULL);
1856
1857 g_free(child->name);
1858 g_free(child);
1859 }
1860
1861 void bdrv_root_unref_child(BdrvChild *child)
1862 {
1863 BlockDriverState *child_bs;
1864
1865 child_bs = child->bs;
1866 bdrv_detach_child(child);
1867 bdrv_unref(child_bs);
1868 }
1869
1870 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1871 {
1872 if (child == NULL) {
1873 return;
1874 }
1875
1876 if (child->bs->inherits_from == parent) {
1877 BdrvChild *c;
1878
1879 /* Remove inherits_from only when the last reference between parent and
1880 * child->bs goes away. */
1881 QLIST_FOREACH(c, &parent->children, next) {
1882 if (c != child && c->bs == child->bs) {
1883 break;
1884 }
1885 }
1886 if (c == NULL) {
1887 child->bs->inherits_from = NULL;
1888 }
1889 }
1890
1891 bdrv_root_unref_child(child);
1892 }
1893
1894
1895 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1896 {
1897 BdrvChild *c;
1898 QLIST_FOREACH(c, &bs->parents, next_parent) {
1899 if (c->role->change_media) {
1900 c->role->change_media(c, load);
1901 }
1902 }
1903 }
1904
1905 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1906 {
1907 BdrvChild *c;
1908 QLIST_FOREACH(c, &bs->parents, next_parent) {
1909 if (c->role->resize) {
1910 c->role->resize(c);
1911 }
1912 }
1913 }
1914
1915 /*
1916 * Sets the backing file link of a BDS. A new reference is created; callers
1917 * which don't need their own reference any more must call bdrv_unref().
1918 */
1919 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
1920 Error **errp)
1921 {
1922 if (backing_hd) {
1923 bdrv_ref(backing_hd);
1924 }
1925
1926 if (bs->backing) {
1927 bdrv_unref_child(bs, bs->backing);
1928 }
1929
1930 if (!backing_hd) {
1931 bs->backing = NULL;
1932 goto out;
1933 }
1934
1935 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
1936 errp);
1937 if (!bs->backing) {
1938 bdrv_unref(backing_hd);
1939 }
1940
1941 bdrv_refresh_filename(bs);
1942
1943 out:
1944 bdrv_refresh_limits(bs, NULL);
1945 }
1946
1947 /*
1948 * Opens the backing file for a BlockDriverState if not yet open
1949 *
1950 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1951 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1952 * itself, all options starting with "${bdref_key}." are considered part of the
1953 * BlockdevRef.
1954 *
1955 * TODO Can this be unified with bdrv_open_image()?
1956 */
1957 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1958 const char *bdref_key, Error **errp)
1959 {
1960 char *backing_filename = g_malloc0(PATH_MAX);
1961 char *bdref_key_dot;
1962 const char *reference = NULL;
1963 int ret = 0;
1964 BlockDriverState *backing_hd;
1965 QDict *options;
1966 QDict *tmp_parent_options = NULL;
1967 Error *local_err = NULL;
1968
1969 if (bs->backing != NULL) {
1970 goto free_exit;
1971 }
1972
1973 /* NULL means an empty set of options */
1974 if (parent_options == NULL) {
1975 tmp_parent_options = qdict_new();
1976 parent_options = tmp_parent_options;
1977 }
1978
1979 bs->open_flags &= ~BDRV_O_NO_BACKING;
1980
1981 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1982 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
1983 g_free(bdref_key_dot);
1984
1985 reference = qdict_get_try_str(parent_options, bdref_key);
1986 if (reference || qdict_haskey(options, "file.filename")) {
1987 backing_filename[0] = '\0';
1988 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1989 QDECREF(options);
1990 goto free_exit;
1991 } else {
1992 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
1993 &local_err);
1994 if (local_err) {
1995 ret = -EINVAL;
1996 error_propagate(errp, local_err);
1997 QDECREF(options);
1998 goto free_exit;
1999 }
2000 }
2001
2002 if (!bs->drv || !bs->drv->supports_backing) {
2003 ret = -EINVAL;
2004 error_setg(errp, "Driver doesn't support backing files");
2005 QDECREF(options);
2006 goto free_exit;
2007 }
2008
2009 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2010 qdict_put(options, "driver", qstring_from_str(bs->backing_format));
2011 }
2012
2013 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2014 reference, options, 0, bs, &child_backing,
2015 errp);
2016 if (!backing_hd) {
2017 bs->open_flags |= BDRV_O_NO_BACKING;
2018 error_prepend(errp, "Could not open backing file: ");
2019 ret = -EINVAL;
2020 goto free_exit;
2021 }
2022
2023 /* Hook up the backing file link; drop our reference, bs owns the
2024 * backing_hd reference now */
2025 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2026 bdrv_unref(backing_hd);
2027 if (local_err) {
2028 ret = -EINVAL;
2029 goto free_exit;
2030 }
2031
2032 qdict_del(parent_options, bdref_key);
2033
2034 free_exit:
2035 g_free(backing_filename);
2036 QDECREF(tmp_parent_options);
2037 return ret;
2038 }
2039
2040 static BlockDriverState *
2041 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2042 BlockDriverState *parent, const BdrvChildRole *child_role,
2043 bool allow_none, Error **errp)
2044 {
2045 BlockDriverState *bs = NULL;
2046 QDict *image_options;
2047 char *bdref_key_dot;
2048 const char *reference;
2049
2050 assert(child_role != NULL);
2051
2052 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2053 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2054 g_free(bdref_key_dot);
2055
2056 reference = qdict_get_try_str(options, bdref_key);
2057 if (!filename && !reference && !qdict_size(image_options)) {
2058 if (!allow_none) {
2059 error_setg(errp, "A block device must be specified for \"%s\"",
2060 bdref_key);
2061 }
2062 QDECREF(image_options);
2063 goto done;
2064 }
2065
2066 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2067 parent, child_role, errp);
2068 if (!bs) {
2069 goto done;
2070 }
2071
2072 done:
2073 qdict_del(options, bdref_key);
2074 return bs;
2075 }
2076
2077 /*
2078 * Opens a disk image whose options are given as BlockdevRef in another block
2079 * device's options.
2080 *
2081 * If allow_none is true, no image will be opened if filename is false and no
2082 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2083 *
2084 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2085 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2086 * itself, all options starting with "${bdref_key}." are considered part of the
2087 * BlockdevRef.
2088 *
2089 * The BlockdevRef will be removed from the options QDict.
2090 */
2091 BdrvChild *bdrv_open_child(const char *filename,
2092 QDict *options, const char *bdref_key,
2093 BlockDriverState *parent,
2094 const BdrvChildRole *child_role,
2095 bool allow_none, Error **errp)
2096 {
2097 BdrvChild *c;
2098 BlockDriverState *bs;
2099
2100 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2101 allow_none, errp);
2102 if (bs == NULL) {
2103 return NULL;
2104 }
2105
2106 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2107 if (!c) {
2108 bdrv_unref(bs);
2109 return NULL;
2110 }
2111
2112 return c;
2113 }
2114
2115 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2116 int flags,
2117 QDict *snapshot_options,
2118 Error **errp)
2119 {
2120 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2121 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2122 int64_t total_size;
2123 QemuOpts *opts = NULL;
2124 BlockDriverState *bs_snapshot;
2125 Error *local_err = NULL;
2126 int ret;
2127
2128 /* if snapshot, we create a temporary backing file and open it
2129 instead of opening 'filename' directly */
2130
2131 /* Get the required size from the image */
2132 total_size = bdrv_getlength(bs);
2133 if (total_size < 0) {
2134 error_setg_errno(errp, -total_size, "Could not get image size");
2135 goto out;
2136 }
2137
2138 /* Create the temporary image */
2139 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2140 if (ret < 0) {
2141 error_setg_errno(errp, -ret, "Could not get temporary filename");
2142 goto out;
2143 }
2144
2145 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2146 &error_abort);
2147 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2148 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2149 qemu_opts_del(opts);
2150 if (ret < 0) {
2151 error_prepend(errp, "Could not create temporary overlay '%s': ",
2152 tmp_filename);
2153 goto out;
2154 }
2155
2156 /* Prepare options QDict for the temporary file */
2157 qdict_put(snapshot_options, "file.driver",
2158 qstring_from_str("file"));
2159 qdict_put(snapshot_options, "file.filename",
2160 qstring_from_str(tmp_filename));
2161 qdict_put(snapshot_options, "driver",
2162 qstring_from_str("qcow2"));
2163
2164 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2165 snapshot_options = NULL;
2166 if (!bs_snapshot) {
2167 ret = -EINVAL;
2168 goto out;
2169 }
2170
2171 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
2172 * call bdrv_unref() on it), so in order to be able to return one, we have
2173 * to increase bs_snapshot's refcount here */
2174 bdrv_ref(bs_snapshot);
2175 bdrv_append(bs_snapshot, bs, &local_err);
2176 if (local_err) {
2177 error_propagate(errp, local_err);
2178 ret = -EINVAL;
2179 goto out;
2180 }
2181
2182 g_free(tmp_filename);
2183 return bs_snapshot;
2184
2185 out:
2186 QDECREF(snapshot_options);
2187 g_free(tmp_filename);
2188 return NULL;
2189 }
2190
2191 /*
2192 * Opens a disk image (raw, qcow2, vmdk, ...)
2193 *
2194 * options is a QDict of options to pass to the block drivers, or NULL for an
2195 * empty set of options. The reference to the QDict belongs to the block layer
2196 * after the call (even on failure), so if the caller intends to reuse the
2197 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2198 *
2199 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2200 * If it is not NULL, the referenced BDS will be reused.
2201 *
2202 * The reference parameter may be used to specify an existing block device which
2203 * should be opened. If specified, neither options nor a filename may be given,
2204 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2205 */
2206 static BlockDriverState *bdrv_open_inherit(const char *filename,
2207 const char *reference,
2208 QDict *options, int flags,
2209 BlockDriverState *parent,
2210 const BdrvChildRole *child_role,
2211 Error **errp)
2212 {
2213 int ret;
2214 BlockBackend *file = NULL;
2215 BlockDriverState *bs;
2216 BlockDriver *drv = NULL;
2217 const char *drvname;
2218 const char *backing;
2219 Error *local_err = NULL;
2220 QDict *snapshot_options = NULL;
2221 int snapshot_flags = 0;
2222
2223 assert(!child_role || !flags);
2224 assert(!child_role == !parent);
2225
2226 if (reference) {
2227 bool options_non_empty = options ? qdict_size(options) : false;
2228 QDECREF(options);
2229
2230 if (filename || options_non_empty) {
2231 error_setg(errp, "Cannot reference an existing block device with "
2232 "additional options or a new filename");
2233 return NULL;
2234 }
2235
2236 bs = bdrv_lookup_bs(reference, reference, errp);
2237 if (!bs) {
2238 return NULL;
2239 }
2240
2241 bdrv_ref(bs);
2242 return bs;
2243 }
2244
2245 bs = bdrv_new();
2246
2247 /* NULL means an empty set of options */
2248 if (options == NULL) {
2249 options = qdict_new();
2250 }
2251
2252 /* json: syntax counts as explicit options, as if in the QDict */
2253 parse_json_protocol(options, &filename, &local_err);
2254 if (local_err) {
2255 goto fail;
2256 }
2257
2258 bs->explicit_options = qdict_clone_shallow(options);
2259
2260 if (child_role) {
2261 bs->inherits_from = parent;
2262 child_role->inherit_options(&flags, options,
2263 parent->open_flags, parent->options);
2264 }
2265
2266 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2267 if (local_err) {
2268 goto fail;
2269 }
2270
2271 /* Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2272 * FIXME: we're parsing the QDict to avoid having to create a
2273 * QemuOpts just for this, but neither option is optimal. */
2274 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2275 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2276 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2277 } else {
2278 flags &= ~BDRV_O_RDWR;
2279 }
2280
2281 if (flags & BDRV_O_SNAPSHOT) {
2282 snapshot_options = qdict_new();
2283 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2284 flags, options);
2285 /* Let bdrv_backing_options() override "read-only" */
2286 qdict_del(options, BDRV_OPT_READ_ONLY);
2287 bdrv_backing_options(&flags, options, flags, options);
2288 }
2289
2290 bs->open_flags = flags;
2291 bs->options = options;
2292 options = qdict_clone_shallow(options);
2293
2294 /* Find the right image format driver */
2295 drvname = qdict_get_try_str(options, "driver");
2296 if (drvname) {
2297 drv = bdrv_find_format(drvname);
2298 if (!drv) {
2299 error_setg(errp, "Unknown driver: '%s'", drvname);
2300 goto fail;
2301 }
2302 }
2303
2304 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2305
2306 backing = qdict_get_try_str(options, "backing");
2307 if (backing && *backing == '\0') {
2308 flags |= BDRV_O_NO_BACKING;
2309 qdict_del(options, "backing");
2310 }
2311
2312 /* Open image file without format layer. This BlockBackend is only used for
2313 * probing, the block drivers will do their own bdrv_open_child() for the
2314 * same BDS, which is why we put the node name back into options. */
2315 if ((flags & BDRV_O_PROTOCOL) == 0) {
2316 BlockDriverState *file_bs;
2317
2318 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2319 &child_file, true, &local_err);
2320 if (local_err) {
2321 goto fail;
2322 }
2323 if (file_bs != NULL) {
2324 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2325 blk_insert_bs(file, file_bs, &local_err);
2326 bdrv_unref(file_bs);
2327 if (local_err) {
2328 goto fail;
2329 }
2330
2331 qdict_put(options, "file",
2332 qstring_from_str(bdrv_get_node_name(file_bs)));
2333 }
2334 }
2335
2336 /* Image format probing */
2337 bs->probed = !drv;
2338 if (!drv && file) {
2339 ret = find_image_format(file, filename, &drv, &local_err);
2340 if (ret < 0) {
2341 goto fail;
2342 }
2343 /*
2344 * This option update would logically belong in bdrv_fill_options(),
2345 * but we first need to open bs->file for the probing to work, while
2346 * opening bs->file already requires the (mostly) final set of options
2347 * so that cache mode etc. can be inherited.
2348 *
2349 * Adding the driver later is somewhat ugly, but it's not an option
2350 * that would ever be inherited, so it's correct. We just need to make
2351 * sure to update both bs->options (which has the full effective
2352 * options for bs) and options (which has file.* already removed).
2353 */
2354 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
2355 qdict_put(options, "driver", qstring_from_str(drv->format_name));
2356 } else if (!drv) {
2357 error_setg(errp, "Must specify either driver or file");
2358 goto fail;
2359 }
2360
2361 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2362 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2363 /* file must be NULL if a protocol BDS is about to be created
2364 * (the inverse results in an error message from bdrv_open_common()) */
2365 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2366
2367 /* Open the image */
2368 ret = bdrv_open_common(bs, file, options, &local_err);
2369 if (ret < 0) {
2370 goto fail;
2371 }
2372
2373 if (file) {
2374 blk_unref(file);
2375 file = NULL;
2376 }
2377
2378 /* If there is a backing file, use it */
2379 if ((flags & BDRV_O_NO_BACKING) == 0) {
2380 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2381 if (ret < 0) {
2382 goto close_and_fail;
2383 }
2384 }
2385
2386 bdrv_refresh_filename(bs);
2387
2388 /* Check if any unknown options were used */
2389 if (qdict_size(options) != 0) {
2390 const QDictEntry *entry = qdict_first(options);
2391 if (flags & BDRV_O_PROTOCOL) {
2392 error_setg(errp, "Block protocol '%s' doesn't support the option "
2393 "'%s'", drv->format_name, entry->key);
2394 } else {
2395 error_setg(errp,
2396 "Block format '%s' does not support the option '%s'",
2397 drv->format_name, entry->key);
2398 }
2399
2400 goto close_and_fail;
2401 }
2402
2403 if (!bdrv_key_required(bs)) {
2404 bdrv_parent_cb_change_media(bs, true);
2405 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2406 && !runstate_check(RUN_STATE_INMIGRATE)
2407 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2408 error_setg(errp,
2409 "Guest must be stopped for opening of encrypted image");
2410 goto close_and_fail;
2411 }
2412
2413 QDECREF(options);
2414
2415 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2416 * temporary snapshot afterwards. */
2417 if (snapshot_flags) {
2418 BlockDriverState *snapshot_bs;
2419 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2420 snapshot_options, &local_err);
2421 snapshot_options = NULL;
2422 if (local_err) {
2423 goto close_and_fail;
2424 }
2425 /* We are not going to return bs but the overlay on top of it
2426 * (snapshot_bs); thus, we have to drop the strong reference to bs
2427 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2428 * though, because the overlay still has a reference to it. */
2429 bdrv_unref(bs);
2430 bs = snapshot_bs;
2431 }
2432
2433 return bs;
2434
2435 fail:
2436 blk_unref(file);
2437 if (bs->file != NULL) {
2438 bdrv_unref_child(bs, bs->file);
2439 }
2440 QDECREF(snapshot_options);
2441 QDECREF(bs->explicit_options);
2442 QDECREF(bs->options);
2443 QDECREF(options);
2444 bs->options = NULL;
2445 bdrv_unref(bs);
2446 error_propagate(errp, local_err);
2447 return NULL;
2448
2449 close_and_fail:
2450 bdrv_unref(bs);
2451 QDECREF(snapshot_options);
2452 QDECREF(options);
2453 error_propagate(errp, local_err);
2454 return NULL;
2455 }
2456
2457 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2458 QDict *options, int flags, Error **errp)
2459 {
2460 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2461 NULL, errp);
2462 }
2463
2464 typedef struct BlockReopenQueueEntry {
2465 bool prepared;
2466 BDRVReopenState state;
2467 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2468 } BlockReopenQueueEntry;
2469
2470 /*
2471 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2472 * reopen of multiple devices.
2473 *
2474 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2475 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2476 * be created and initialized. This newly created BlockReopenQueue should be
2477 * passed back in for subsequent calls that are intended to be of the same
2478 * atomic 'set'.
2479 *
2480 * bs is the BlockDriverState to add to the reopen queue.
2481 *
2482 * options contains the changed options for the associated bs
2483 * (the BlockReopenQueue takes ownership)
2484 *
2485 * flags contains the open flags for the associated bs
2486 *
2487 * returns a pointer to bs_queue, which is either the newly allocated
2488 * bs_queue, or the existing bs_queue being used.
2489 *
2490 */
2491 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2492 BlockDriverState *bs,
2493 QDict *options,
2494 int flags,
2495 const BdrvChildRole *role,
2496 QDict *parent_options,
2497 int parent_flags)
2498 {
2499 assert(bs != NULL);
2500
2501 BlockReopenQueueEntry *bs_entry;
2502 BdrvChild *child;
2503 QDict *old_options, *explicit_options;
2504
2505 if (bs_queue == NULL) {
2506 bs_queue = g_new0(BlockReopenQueue, 1);
2507 QSIMPLEQ_INIT(bs_queue);
2508 }
2509
2510 if (!options) {
2511 options = qdict_new();
2512 }
2513
2514 /* Check if this BlockDriverState is already in the queue */
2515 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2516 if (bs == bs_entry->state.bs) {
2517 break;
2518 }
2519 }
2520
2521 /*
2522 * Precedence of options:
2523 * 1. Explicitly passed in options (highest)
2524 * 2. Set in flags (only for top level)
2525 * 3. Retained from explicitly set options of bs
2526 * 4. Inherited from parent node
2527 * 5. Retained from effective options of bs
2528 */
2529
2530 if (!parent_options) {
2531 /*
2532 * Any setting represented by flags is always updated. If the
2533 * corresponding QDict option is set, it takes precedence. Otherwise
2534 * the flag is translated into a QDict option. The old setting of bs is
2535 * not considered.
2536 */
2537 update_options_from_flags(options, flags);
2538 }
2539
2540 /* Old explicitly set values (don't overwrite by inherited value) */
2541 if (bs_entry) {
2542 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2543 } else {
2544 old_options = qdict_clone_shallow(bs->explicit_options);
2545 }
2546 bdrv_join_options(bs, options, old_options);
2547 QDECREF(old_options);
2548
2549 explicit_options = qdict_clone_shallow(options);
2550
2551 /* Inherit from parent node */
2552 if (parent_options) {
2553 assert(!flags);
2554 role->inherit_options(&flags, options, parent_flags, parent_options);
2555 }
2556
2557 /* Old values are used for options that aren't set yet */
2558 old_options = qdict_clone_shallow(bs->options);
2559 bdrv_join_options(bs, options, old_options);
2560 QDECREF(old_options);
2561
2562 /* bdrv_open() masks this flag out */
2563 flags &= ~BDRV_O_PROTOCOL;
2564
2565 QLIST_FOREACH(child, &bs->children, next) {
2566 QDict *new_child_options;
2567 char *child_key_dot;
2568
2569 /* reopen can only change the options of block devices that were
2570 * implicitly created and inherited options. For other (referenced)
2571 * block devices, a syntax like "backing.foo" results in an error. */
2572 if (child->bs->inherits_from != bs) {
2573 continue;
2574 }
2575
2576 child_key_dot = g_strdup_printf("%s.", child->name);
2577 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2578 g_free(child_key_dot);
2579
2580 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2581 child->role, options, flags);
2582 }
2583
2584 if (!bs_entry) {
2585 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2586 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2587 } else {
2588 QDECREF(bs_entry->state.options);
2589 QDECREF(bs_entry->state.explicit_options);
2590 }
2591
2592 bs_entry->state.bs = bs;
2593 bs_entry->state.options = options;
2594 bs_entry->state.explicit_options = explicit_options;
2595 bs_entry->state.flags = flags;
2596
2597 return bs_queue;
2598 }
2599
2600 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2601 BlockDriverState *bs,
2602 QDict *options, int flags)
2603 {
2604 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2605 NULL, NULL, 0);
2606 }
2607
2608 /*
2609 * Reopen multiple BlockDriverStates atomically & transactionally.
2610 *
2611 * The queue passed in (bs_queue) must have been built up previous
2612 * via bdrv_reopen_queue().
2613 *
2614 * Reopens all BDS specified in the queue, with the appropriate
2615 * flags. All devices are prepared for reopen, and failure of any
2616 * device will cause all device changes to be abandonded, and intermediate
2617 * data cleaned up.
2618 *
2619 * If all devices prepare successfully, then the changes are committed
2620 * to all devices.
2621 *
2622 */
2623 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2624 {
2625 int ret = -1;
2626 BlockReopenQueueEntry *bs_entry, *next;
2627 Error *local_err = NULL;
2628
2629 assert(bs_queue != NULL);
2630
2631 aio_context_release(ctx);
2632 bdrv_drain_all_begin();
2633 aio_context_acquire(ctx);
2634
2635 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2636 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2637 error_propagate(errp, local_err);
2638 goto cleanup;
2639 }
2640 bs_entry->prepared = true;
2641 }
2642
2643 /* If we reach this point, we have success and just need to apply the
2644 * changes
2645 */
2646 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2647 bdrv_reopen_commit(&bs_entry->state);
2648 }
2649
2650 ret = 0;
2651
2652 cleanup:
2653 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2654 if (ret && bs_entry->prepared) {
2655 bdrv_reopen_abort(&bs_entry->state);
2656 } else if (ret) {
2657 QDECREF(bs_entry->state.explicit_options);
2658 }
2659 QDECREF(bs_entry->state.options);
2660 g_free(bs_entry);
2661 }
2662 g_free(bs_queue);
2663
2664 bdrv_drain_all_end();
2665
2666 return ret;
2667 }
2668
2669
2670 /* Reopen a single BlockDriverState with the specified flags. */
2671 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2672 {
2673 int ret = -1;
2674 Error *local_err = NULL;
2675 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2676
2677 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2678 if (local_err != NULL) {
2679 error_propagate(errp, local_err);
2680 }
2681 return ret;
2682 }
2683
2684
2685 /*
2686 * Prepares a BlockDriverState for reopen. All changes are staged in the
2687 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2688 * the block driver layer .bdrv_reopen_prepare()
2689 *
2690 * bs is the BlockDriverState to reopen
2691 * flags are the new open flags
2692 * queue is the reopen queue
2693 *
2694 * Returns 0 on success, non-zero on error. On error errp will be set
2695 * as well.
2696 *
2697 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2698 * It is the responsibility of the caller to then call the abort() or
2699 * commit() for any other BDS that have been left in a prepare() state
2700 *
2701 */
2702 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2703 Error **errp)
2704 {
2705 int ret = -1;
2706 Error *local_err = NULL;
2707 BlockDriver *drv;
2708 QemuOpts *opts;
2709 const char *value;
2710
2711 assert(reopen_state != NULL);
2712 assert(reopen_state->bs->drv != NULL);
2713 drv = reopen_state->bs->drv;
2714
2715 /* Process generic block layer options */
2716 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2717 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2718 if (local_err) {
2719 error_propagate(errp, local_err);
2720 ret = -EINVAL;
2721 goto error;
2722 }
2723
2724 update_flags_from_options(&reopen_state->flags, opts);
2725
2726 /* node-name and driver must be unchanged. Put them back into the QDict, so
2727 * that they are checked at the end of this function. */
2728 value = qemu_opt_get(opts, "node-name");
2729 if (value) {
2730 qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2731 }
2732
2733 value = qemu_opt_get(opts, "driver");
2734 if (value) {
2735 qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2736 }
2737
2738 /* if we are to stay read-only, do not allow permission change
2739 * to r/w */
2740 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2741 reopen_state->flags & BDRV_O_RDWR) {
2742 error_setg(errp, "Node '%s' is read only",
2743 bdrv_get_device_or_node_name(reopen_state->bs));
2744 goto error;
2745 }
2746
2747
2748 ret = bdrv_flush(reopen_state->bs);
2749 if (ret) {
2750 error_setg_errno(errp, -ret, "Error flushing drive");
2751 goto error;
2752 }
2753
2754 if (drv->bdrv_reopen_prepare) {
2755 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2756 if (ret) {
2757 if (local_err != NULL) {
2758 error_propagate(errp, local_err);
2759 } else {
2760 error_setg(errp, "failed while preparing to reopen image '%s'",
2761 reopen_state->bs->filename);
2762 }
2763 goto error;
2764 }
2765 } else {
2766 /* It is currently mandatory to have a bdrv_reopen_prepare()
2767 * handler for each supported drv. */
2768 error_setg(errp, "Block format '%s' used by node '%s' "
2769 "does not support reopening files", drv->format_name,
2770 bdrv_get_device_or_node_name(reopen_state->bs));
2771 ret = -1;
2772 goto error;
2773 }
2774
2775 /* Options that are not handled are only okay if they are unchanged
2776 * compared to the old state. It is expected that some options are only
2777 * used for the initial open, but not reopen (e.g. filename) */
2778 if (qdict_size(reopen_state->options)) {
2779 const QDictEntry *entry = qdict_first(reopen_state->options);
2780
2781 do {
2782 QString *new_obj = qobject_to_qstring(entry->value);
2783 const char *new = qstring_get_str(new_obj);
2784 const char *old = qdict_get_try_str(reopen_state->bs->options,
2785 entry->key);
2786
2787 if (!old || strcmp(new, old)) {
2788 error_setg(errp, "Cannot change the option '%s'", entry->key);
2789 ret = -EINVAL;
2790 goto error;
2791 }
2792 } while ((entry = qdict_next(reopen_state->options, entry)));
2793 }
2794
2795 ret = 0;
2796
2797 error:
2798 qemu_opts_del(opts);
2799 return ret;
2800 }
2801
2802 /*
2803 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2804 * makes them final by swapping the staging BlockDriverState contents into
2805 * the active BlockDriverState contents.
2806 */
2807 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2808 {
2809 BlockDriver *drv;
2810
2811 assert(reopen_state != NULL);
2812 drv = reopen_state->bs->drv;
2813 assert(drv != NULL);
2814
2815 /* If there are any driver level actions to take */
2816 if (drv->bdrv_reopen_commit) {
2817 drv->bdrv_reopen_commit(reopen_state);
2818 }
2819
2820 /* set BDS specific flags now */
2821 QDECREF(reopen_state->bs->explicit_options);
2822
2823 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2824 reopen_state->bs->open_flags = reopen_state->flags;
2825 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2826
2827 bdrv_refresh_limits(reopen_state->bs, NULL);
2828 }
2829
2830 /*
2831 * Abort the reopen, and delete and free the staged changes in
2832 * reopen_state
2833 */
2834 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2835 {
2836 BlockDriver *drv;
2837
2838 assert(reopen_state != NULL);
2839 drv = reopen_state->bs->drv;
2840 assert(drv != NULL);
2841
2842 if (drv->bdrv_reopen_abort) {
2843 drv->bdrv_reopen_abort(reopen_state);
2844 }
2845
2846 QDECREF(reopen_state->explicit_options);
2847 }
2848
2849
2850 static void bdrv_close(BlockDriverState *bs)
2851 {
2852 BdrvAioNotifier *ban, *ban_next;
2853
2854 assert(!bs->job);
2855 assert(!bs->refcnt);
2856
2857 bdrv_drained_begin(bs); /* complete I/O */
2858 bdrv_flush(bs);
2859 bdrv_drain(bs); /* in case flush left pending I/O */
2860
2861 bdrv_release_named_dirty_bitmaps(bs);
2862 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2863
2864 if (bs->drv) {
2865 BdrvChild *child, *next;
2866
2867 bs->drv->bdrv_close(bs);
2868 bs->drv = NULL;
2869
2870 bdrv_set_backing_hd(bs, NULL, &error_abort);
2871
2872 if (bs->file != NULL) {
2873 bdrv_unref_child(bs, bs->file);
2874 bs->file = NULL;
2875 }
2876
2877 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2878 /* TODO Remove bdrv_unref() from drivers' close function and use
2879 * bdrv_unref_child() here */
2880 if (child->bs->inherits_from == bs) {
2881 child->bs->inherits_from = NULL;
2882 }
2883 bdrv_detach_child(child);
2884 }
2885
2886 g_free(bs->opaque);
2887 bs->opaque = NULL;
2888 bs->copy_on_read = 0;
2889 bs->backing_file[0] = '\0';
2890 bs->backing_format[0] = '\0';
2891 bs->total_sectors = 0;
2892 bs->encrypted = false;
2893 bs->valid_key = false;
2894 bs->sg = false;
2895 QDECREF(bs->options);
2896 QDECREF(bs->explicit_options);
2897 bs->options = NULL;
2898 QDECREF(bs->full_open_options);
2899 bs->full_open_options = NULL;
2900 }
2901
2902 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2903 g_free(ban);
2904 }
2905 QLIST_INIT(&bs->aio_notifiers);
2906 bdrv_drained_end(bs);
2907 }
2908
2909 void bdrv_close_all(void)
2910 {
2911 block_job_cancel_sync_all();
2912 nbd_export_close_all();
2913
2914 /* Drop references from requests still in flight, such as canceled block
2915 * jobs whose AIO context has not been polled yet */
2916 bdrv_drain_all();
2917
2918 blk_remove_all_bs();
2919 blockdev_close_all_bdrv_states();
2920
2921 assert(QTAILQ_EMPTY(&all_bdrv_states));
2922 }
2923
2924 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
2925 {
2926 BdrvChild *to_c;
2927
2928 if (c->role->stay_at_node) {
2929 return false;
2930 }
2931
2932 if (c->role == &child_backing) {
2933 /* If @from is a backing file of @to, ignore the child to avoid
2934 * creating a loop. We only want to change the pointer of other
2935 * parents. */
2936 QLIST_FOREACH(to_c, &to->children, next) {
2937 if (to_c == c) {
2938 break;
2939 }
2940 }
2941 if (to_c) {
2942 return false;
2943 }
2944 }
2945
2946 return true;
2947 }
2948
2949 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
2950 Error **errp)
2951 {
2952 BdrvChild *c, *next;
2953 GSList *list = NULL, *p;
2954 uint64_t old_perm, old_shared;
2955 uint64_t perm = 0, shared = BLK_PERM_ALL;
2956 int ret;
2957
2958 assert(!atomic_read(&from->in_flight));
2959 assert(!atomic_read(&to->in_flight));
2960
2961 /* Make sure that @from doesn't go away until we have successfully attached
2962 * all of its parents to @to. */
2963 bdrv_ref(from);
2964
2965 /* Put all parents into @list and calculate their cumulative permissions */
2966 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
2967 if (!should_update_child(c, to)) {
2968 continue;
2969 }
2970 list = g_slist_prepend(list, c);
2971 perm |= c->perm;
2972 shared &= c->shared_perm;
2973 }
2974
2975 /* Check whether the required permissions can be granted on @to, ignoring
2976 * all BdrvChild in @list so that they can't block themselves. */
2977 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
2978 if (ret < 0) {
2979 bdrv_abort_perm_update(to);
2980 goto out;
2981 }
2982
2983 /* Now actually perform the change. We performed the permission check for
2984 * all elements of @list at once, so set the permissions all at once at the
2985 * very end. */
2986 for (p = list; p != NULL; p = p->next) {
2987 c = p->data;
2988
2989 bdrv_ref(to);
2990 bdrv_replace_child_noperm(c, to);
2991 bdrv_unref(from);
2992 }
2993
2994 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
2995 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
2996
2997 out:
2998 g_slist_free(list);
2999 bdrv_unref(from);
3000 }
3001
3002 /*
3003 * Add new bs contents at the top of an image chain while the chain is
3004 * live, while keeping required fields on the top layer.
3005 *
3006 * This will modify the BlockDriverState fields, and swap contents
3007 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3008 *
3009 * bs_new must not be attached to a BlockBackend.
3010 *
3011 * This function does not create any image files.
3012 *
3013 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3014 * that's what the callers commonly need. bs_new will be referenced by the old
3015 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3016 * reference of its own, it must call bdrv_ref().
3017 */
3018 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3019 Error **errp)
3020 {
3021 Error *local_err = NULL;
3022
3023 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3024 if (local_err) {
3025 error_propagate(errp, local_err);
3026 goto out;
3027 }
3028
3029 bdrv_replace_node(bs_top, bs_new, &local_err);
3030 if (local_err) {
3031 error_propagate(errp, local_err);
3032 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3033 goto out;
3034 }
3035
3036 /* bs_new is now referenced by its new parents, we don't need the
3037 * additional reference any more. */
3038 out:
3039 bdrv_unref(bs_new);
3040 }
3041
3042 static void bdrv_delete(BlockDriverState *bs)
3043 {
3044 assert(!bs->job);
3045 assert(bdrv_op_blocker_is_empty(bs));
3046 assert(!bs->refcnt);
3047
3048 bdrv_close(bs);
3049
3050 /* remove from list, if necessary */
3051 if (bs->node_name[0] != '\0') {
3052 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3053 }
3054 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3055
3056 g_free(bs);
3057 }
3058
3059 /*
3060 * Run consistency checks on an image
3061 *
3062 * Returns 0 if the check could be completed (it doesn't mean that the image is
3063 * free of errors) or -errno when an internal error occurred. The results of the
3064 * check are stored in res.
3065 */
3066 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3067 {
3068 if (bs->drv == NULL) {
3069 return -ENOMEDIUM;
3070 }
3071 if (bs->drv->bdrv_check == NULL) {
3072 return -ENOTSUP;
3073 }
3074
3075 memset(res, 0, sizeof(*res));
3076 return bs->drv->bdrv_check(bs, res, fix);
3077 }
3078
3079 /*
3080 * Return values:
3081 * 0 - success
3082 * -EINVAL - backing format specified, but no file
3083 * -ENOSPC - can't update the backing file because no space is left in the
3084 * image file header
3085 * -ENOTSUP - format driver doesn't support changing the backing file
3086 */
3087 int bdrv_change_backing_file(BlockDriverState *bs,
3088 const char *backing_file, const char *backing_fmt)
3089 {
3090 BlockDriver *drv = bs->drv;
3091 int ret;
3092
3093 /* Backing file format doesn't make sense without a backing file */
3094 if (backing_fmt && !backing_file) {
3095 return -EINVAL;
3096 }
3097
3098 if (drv->bdrv_change_backing_file != NULL) {
3099 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3100 } else {
3101 ret = -ENOTSUP;
3102 }
3103
3104 if (ret == 0) {
3105 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3106 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3107 }
3108 return ret;
3109 }
3110
3111 /*
3112 * Finds the image layer in the chain that has 'bs' as its backing file.
3113 *
3114 * active is the current topmost image.
3115 *
3116 * Returns NULL if bs is not found in active's image chain,
3117 * or if active == bs.
3118 *
3119 * Returns the bottommost base image if bs == NULL.
3120 */
3121 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3122 BlockDriverState *bs)
3123 {
3124 while (active && bs != backing_bs(active)) {
3125 active = backing_bs(active);
3126 }
3127
3128 return active;
3129 }
3130
3131 /* Given a BDS, searches for the base layer. */
3132 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3133 {
3134 return bdrv_find_overlay(bs, NULL);
3135 }
3136
3137 /*
3138 * Drops images above 'base' up to and including 'top', and sets the image
3139 * above 'top' to have base as its backing file.
3140 *
3141 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3142 * information in 'bs' can be properly updated.
3143 *
3144 * E.g., this will convert the following chain:
3145 * bottom <- base <- intermediate <- top <- active
3146 *
3147 * to
3148 *
3149 * bottom <- base <- active
3150 *
3151 * It is allowed for bottom==base, in which case it converts:
3152 *
3153 * base <- intermediate <- top <- active
3154 *
3155 * to
3156 *
3157 * base <- active
3158 *
3159 * If backing_file_str is non-NULL, it will be used when modifying top's
3160 * overlay image metadata.
3161 *
3162 * Error conditions:
3163 * if active == top, that is considered an error
3164 *
3165 */
3166 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3167 BlockDriverState *base, const char *backing_file_str)
3168 {
3169 BlockDriverState *new_top_bs = NULL;
3170 Error *local_err = NULL;
3171 int ret = -EIO;
3172
3173 if (!top->drv || !base->drv) {
3174 goto exit;
3175 }
3176
3177 new_top_bs = bdrv_find_overlay(active, top);
3178
3179 if (new_top_bs == NULL) {
3180 /* we could not find the image above 'top', this is an error */
3181 goto exit;
3182 }
3183
3184 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3185 * to do, no intermediate images */
3186 if (backing_bs(new_top_bs) == base) {
3187 ret = 0;
3188 goto exit;
3189 }
3190
3191 /* Make sure that base is in the backing chain of top */
3192 if (!bdrv_chain_contains(top, base)) {
3193 goto exit;
3194 }
3195
3196 /* success - we can delete the intermediate states, and link top->base */
3197 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3198 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3199 base->drv ? base->drv->format_name : "");
3200 if (ret) {
3201 goto exit;
3202 }
3203
3204 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3205 if (local_err) {
3206 ret = -EPERM;
3207 error_report_err(local_err);
3208 goto exit;
3209 }
3210
3211 ret = 0;
3212 exit:
3213 return ret;
3214 }
3215
3216 /**
3217 * Truncate file to 'offset' bytes (needed only for file protocols)
3218 */
3219 int bdrv_truncate(BdrvChild *child, int64_t offset)
3220 {
3221 BlockDriverState *bs = child->bs;
3222 BlockDriver *drv = bs->drv;
3223 int ret;
3224
3225 assert(child->perm & BLK_PERM_RESIZE);
3226
3227 if (!drv)
3228 return -ENOMEDIUM;
3229 if (!drv->bdrv_truncate)
3230 return -ENOTSUP;
3231 if (bs->read_only)
3232 return -EACCES;
3233
3234 ret = drv->bdrv_truncate(bs, offset);
3235 if (ret == 0) {
3236 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3237 bdrv_dirty_bitmap_truncate(bs);
3238 bdrv_parent_cb_resize(bs);
3239 ++bs->write_gen;
3240 }
3241 return ret;
3242 }
3243
3244 /**
3245 * Length of a allocated file in bytes. Sparse files are counted by actual
3246 * allocated space. Return < 0 if error or unknown.
3247 */
3248 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3249 {
3250 BlockDriver *drv = bs->drv;
3251 if (!drv) {
3252 return -ENOMEDIUM;
3253 }
3254 if (drv->bdrv_get_allocated_file_size) {
3255 return drv->bdrv_get_allocated_file_size(bs);
3256 }
3257 if (bs->file) {
3258 return bdrv_get_allocated_file_size(bs->file->bs);
3259 }
3260 return -ENOTSUP;
3261 }
3262
3263 /**
3264 * Return number of sectors on success, -errno on error.
3265 */
3266 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3267 {
3268 BlockDriver *drv = bs->drv;
3269
3270 if (!drv)
3271 return -ENOMEDIUM;
3272
3273 if (drv->has_variable_length) {
3274 int ret = refresh_total_sectors(bs, bs->total_sectors);
3275 if (ret < 0) {
3276 return ret;
3277 }
3278 }
3279 return bs->total_sectors;
3280 }
3281
3282 /**
3283 * Return length in bytes on success, -errno on error.
3284 * The length is always a multiple of BDRV_SECTOR_SIZE.
3285 */
3286 int64_t bdrv_getlength(BlockDriverState *bs)
3287 {
3288 int64_t ret = bdrv_nb_sectors(bs);
3289
3290 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3291 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3292 }
3293
3294 /* return 0 as number of sectors if no device present or error */
3295 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3296 {
3297 int64_t nb_sectors = bdrv_nb_sectors(bs);
3298
3299 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3300 }
3301
3302 bool bdrv_is_read_only(BlockDriverState *bs)
3303 {
3304 return bs->read_only;
3305 }
3306
3307 bool bdrv_is_sg(BlockDriverState *bs)
3308 {
3309 return bs->sg;
3310 }
3311
3312 bool bdrv_is_encrypted(BlockDriverState *bs)
3313 {
3314 if (bs->backing && bs->backing->bs->encrypted) {
3315 return true;
3316 }
3317 return bs->encrypted;
3318 }
3319
3320 bool bdrv_key_required(BlockDriverState *bs)
3321 {
3322 BdrvChild *backing = bs->backing;
3323
3324 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3325 return true;
3326 }
3327 return (bs->encrypted && !bs->valid_key);
3328 }
3329
3330 int bdrv_set_key(BlockDriverState *bs, const char *key)
3331 {
3332 int ret;
3333 if (bs->backing && bs->backing->bs->encrypted) {
3334 ret = bdrv_set_key(bs->backing->bs, key);
3335 if (ret < 0)
3336 return ret;
3337 if (!bs->encrypted)
3338 return 0;
3339 }
3340 if (!bs->encrypted) {
3341 return -EINVAL;
3342 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3343 return -ENOMEDIUM;
3344 }
3345 ret = bs->drv->bdrv_set_key(bs, key);
3346 if (ret < 0) {
3347 bs->valid_key = false;
3348 } else if (!bs->valid_key) {
3349 /* call the change callback now, we skipped it on open */
3350 bs->valid_key = true;
3351 bdrv_parent_cb_change_media(bs, true);
3352 }
3353 return ret;
3354 }
3355
3356 /*
3357 * Provide an encryption key for @bs.
3358 * If @key is non-null:
3359 * If @bs is not encrypted, fail.
3360 * Else if the key is invalid, fail.
3361 * Else set @bs's key to @key, replacing the existing key, if any.
3362 * If @key is null:
3363 * If @bs is encrypted and still lacks a key, fail.
3364 * Else do nothing.
3365 * On failure, store an error object through @errp if non-null.
3366 */
3367 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3368 {
3369 if (key) {
3370 if (!bdrv_is_encrypted(bs)) {
3371 error_setg(errp, "Node '%s' is not encrypted",
3372 bdrv_get_device_or_node_name(bs));
3373 } else if (bdrv_set_key(bs, key) < 0) {
3374 error_setg(errp, QERR_INVALID_PASSWORD);
3375 }
3376 } else {
3377 if (bdrv_key_required(bs)) {
3378 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3379 "'%s' (%s) is encrypted",
3380 bdrv_get_device_or_node_name(bs),
3381 bdrv_get_encrypted_filename(bs));
3382 }
3383 }
3384 }
3385
3386 const char *bdrv_get_format_name(BlockDriverState *bs)
3387 {
3388 return bs->drv ? bs->drv->format_name : NULL;
3389 }
3390
3391 static int qsort_strcmp(const void *a, const void *b)
3392 {
3393 return strcmp(*(char *const *)a, *(char *const *)b);
3394 }
3395
3396 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3397 void *opaque)
3398 {
3399 BlockDriver *drv;
3400 int count = 0;
3401 int i;
3402 const char **formats = NULL;
3403
3404 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3405 if (drv->format_name) {
3406 bool found = false;
3407 int i = count;
3408 while (formats && i && !found) {
3409 found = !strcmp(formats[--i], drv->format_name);
3410 }
3411
3412 if (!found) {
3413 formats = g_renew(const char *, formats, count + 1);
3414 formats[count++] = drv->format_name;
3415 }
3416 }
3417 }
3418
3419 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3420 const char *format_name = block_driver_modules[i].format_name;
3421
3422 if (format_name) {
3423 bool found = false;
3424 int j = count;
3425
3426 while (formats && j && !found) {
3427 found = !strcmp(formats[--j], format_name);
3428 }
3429
3430 if (!found) {
3431 formats = g_renew(const char *, formats, count + 1);
3432 formats[count++] = format_name;
3433 }
3434 }
3435 }
3436
3437 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3438
3439 for (i = 0; i < count; i++) {
3440 it(opaque, formats[i]);
3441 }
3442
3443 g_free(formats);
3444 }
3445
3446 /* This function is to find a node in the bs graph */
3447 BlockDriverState *bdrv_find_node(const char *node_name)
3448 {
3449 BlockDriverState *bs;
3450
3451 assert(node_name);
3452
3453 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3454 if (!strcmp(node_name, bs->node_name)) {
3455 return bs;
3456 }
3457 }
3458 return NULL;
3459 }
3460
3461 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3462 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3463 {
3464 BlockDeviceInfoList *list, *entry;
3465 BlockDriverState *bs;
3466
3467 list = NULL;
3468 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3469 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3470 if (!info) {
3471 qapi_free_BlockDeviceInfoList(list);
3472 return NULL;
3473 }
3474 entry = g_malloc0(sizeof(*entry));
3475 entry->value = info;
3476 entry->next = list;
3477 list = entry;
3478 }
3479
3480 return list;
3481 }
3482
3483 BlockDriverState *bdrv_lookup_bs(const char *device,
3484 const char *node_name,
3485 Error **errp)
3486 {
3487 BlockBackend *blk;
3488 BlockDriverState *bs;
3489
3490 if (device) {
3491 blk = blk_by_name(device);
3492
3493 if (blk) {
3494 bs = blk_bs(blk);
3495 if (!bs) {
3496 error_setg(errp, "Device '%s' has no medium", device);
3497 }
3498
3499 return bs;
3500 }
3501 }
3502
3503 if (node_name) {
3504 bs = bdrv_find_node(node_name);
3505
3506 if (bs) {
3507 return bs;
3508 }
3509 }
3510
3511 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3512 device ? device : "",
3513 node_name ? node_name : "");
3514 return NULL;
3515 }
3516
3517 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3518 * return false. If either argument is NULL, return false. */
3519 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3520 {
3521 while (top && top != base) {
3522 top = backing_bs(top);
3523 }
3524
3525 return top != NULL;
3526 }
3527
3528 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3529 {
3530 if (!bs) {
3531 return QTAILQ_FIRST(&graph_bdrv_states);
3532 }
3533 return QTAILQ_NEXT(bs, node_list);
3534 }
3535
3536 const char *bdrv_get_node_name(const BlockDriverState *bs)
3537 {
3538 return bs->node_name;
3539 }
3540
3541 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3542 {
3543 BdrvChild *c;
3544 const char *name;
3545
3546 /* If multiple parents have a name, just pick the first one. */
3547 QLIST_FOREACH(c, &bs->parents, next_parent) {
3548 if (c->role->get_name) {
3549 name = c->role->get_name(c);
3550 if (name && *name) {
3551 return name;
3552 }
3553 }
3554 }
3555
3556 return NULL;
3557 }
3558
3559 /* TODO check what callers really want: bs->node_name or blk_name() */
3560 const char *bdrv_get_device_name(const BlockDriverState *bs)
3561 {
3562 return bdrv_get_parent_name(bs) ?: "";
3563 }
3564
3565 /* This can be used to identify nodes that might not have a device
3566 * name associated. Since node and device names live in the same
3567 * namespace, the result is unambiguous. The exception is if both are
3568 * absent, then this returns an empty (non-null) string. */
3569 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3570 {
3571 return bdrv_get_parent_name(bs) ?: bs->node_name;
3572 }
3573
3574 int bdrv_get_flags(BlockDriverState *bs)
3575 {
3576 return bs->open_flags;
3577 }
3578
3579 int bdrv_has_zero_init_1(BlockDriverState *bs)
3580 {