sm501: Add missing arbitration control register
[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 /*
1161 * Caution: while qdict_get_try_str() is fine, getting
1162 * non-string types would require more care. When @options
1163 * come from -blockdev or blockdev_add, its members are typed
1164 * according to the QAPI schema, but when they come from
1165 * -drive, they're all QString.
1166 */
1167 filename = qdict_get_try_str(options, "filename");
1168 }
1169
1170 if (drv->bdrv_needs_filename && !filename) {
1171 error_setg(errp, "The '%s' block driver requires a file name",
1172 drv->format_name);
1173 ret = -EINVAL;
1174 goto fail_opts;
1175 }
1176
1177 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1178 drv->format_name);
1179
1180 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1181
1182 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1183 error_setg(errp,
1184 !bs->read_only && bdrv_is_whitelisted(drv, true)
1185 ? "Driver '%s' can only be used for read-only devices"
1186 : "Driver '%s' is not whitelisted",
1187 drv->format_name);
1188 ret = -ENOTSUP;
1189 goto fail_opts;
1190 }
1191
1192 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1193 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1194 if (!bs->read_only) {
1195 bdrv_enable_copy_on_read(bs);
1196 } else {
1197 error_setg(errp, "Can't use copy-on-read on read-only device");
1198 ret = -EINVAL;
1199 goto fail_opts;
1200 }
1201 }
1202
1203 discard = qemu_opt_get(opts, "discard");
1204 if (discard != NULL) {
1205 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1206 error_setg(errp, "Invalid discard option");
1207 ret = -EINVAL;
1208 goto fail_opts;
1209 }
1210 }
1211
1212 detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1213 if (detect_zeroes) {
1214 BlockdevDetectZeroesOptions value =
1215 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1216 detect_zeroes,
1217 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1218 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1219 &local_err);
1220 if (local_err) {
1221 error_propagate(errp, local_err);
1222 ret = -EINVAL;
1223 goto fail_opts;
1224 }
1225
1226 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1227 !(bs->open_flags & BDRV_O_UNMAP))
1228 {
1229 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1230 "without setting discard operation to unmap");
1231 ret = -EINVAL;
1232 goto fail_opts;
1233 }
1234
1235 bs->detect_zeroes = value;
1236 }
1237
1238 if (filename != NULL) {
1239 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1240 } else {
1241 bs->filename[0] = '\0';
1242 }
1243 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1244
1245 /* Open the image, either directly or using a protocol */
1246 open_flags = bdrv_open_flags(bs, bs->open_flags);
1247 node_name = qemu_opt_get(opts, "node-name");
1248
1249 assert(!drv->bdrv_file_open || file == NULL);
1250 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1251 if (ret < 0) {
1252 goto fail_opts;
1253 }
1254
1255 qemu_opts_del(opts);
1256 return 0;
1257
1258 fail_opts:
1259 qemu_opts_del(opts);
1260 return ret;
1261 }
1262
1263 static QDict *parse_json_filename(const char *filename, Error **errp)
1264 {
1265 QObject *options_obj;
1266 QDict *options;
1267 int ret;
1268
1269 ret = strstart(filename, "json:", &filename);
1270 assert(ret);
1271
1272 options_obj = qobject_from_json(filename, errp);
1273 if (!options_obj) {
1274 /* Work around qobject_from_json() lossage TODO fix that */
1275 if (errp && !*errp) {
1276 error_setg(errp, "Could not parse the JSON options");
1277 return NULL;
1278 }
1279 error_prepend(errp, "Could not parse the JSON options: ");
1280 return NULL;
1281 }
1282
1283 options = qobject_to_qdict(options_obj);
1284 if (!options) {
1285 qobject_decref(options_obj);
1286 error_setg(errp, "Invalid JSON object given");
1287 return NULL;
1288 }
1289
1290 qdict_flatten(options);
1291
1292 return options;
1293 }
1294
1295 static void parse_json_protocol(QDict *options, const char **pfilename,
1296 Error **errp)
1297 {
1298 QDict *json_options;
1299 Error *local_err = NULL;
1300
1301 /* Parse json: pseudo-protocol */
1302 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1303 return;
1304 }
1305
1306 json_options = parse_json_filename(*pfilename, &local_err);
1307 if (local_err) {
1308 error_propagate(errp, local_err);
1309 return;
1310 }
1311
1312 /* Options given in the filename have lower priority than options
1313 * specified directly */
1314 qdict_join(options, json_options, false);
1315 QDECREF(json_options);
1316 *pfilename = NULL;
1317 }
1318
1319 /*
1320 * Fills in default options for opening images and converts the legacy
1321 * filename/flags pair to option QDict entries.
1322 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1323 * block driver has been specified explicitly.
1324 */
1325 static int bdrv_fill_options(QDict **options, const char *filename,
1326 int *flags, Error **errp)
1327 {
1328 const char *drvname;
1329 bool protocol = *flags & BDRV_O_PROTOCOL;
1330 bool parse_filename = false;
1331 BlockDriver *drv = NULL;
1332 Error *local_err = NULL;
1333
1334 /*
1335 * Caution: while qdict_get_try_str() is fine, getting non-string
1336 * types would require more care. When @options come from
1337 * -blockdev or blockdev_add, its members are typed according to
1338 * the QAPI schema, but when they come from -drive, they're all
1339 * QString.
1340 */
1341 drvname = qdict_get_try_str(*options, "driver");
1342 if (drvname) {
1343 drv = bdrv_find_format(drvname);
1344 if (!drv) {
1345 error_setg(errp, "Unknown driver '%s'", drvname);
1346 return -ENOENT;
1347 }
1348 /* If the user has explicitly specified the driver, this choice should
1349 * override the BDRV_O_PROTOCOL flag */
1350 protocol = drv->bdrv_file_open;
1351 }
1352
1353 if (protocol) {
1354 *flags |= BDRV_O_PROTOCOL;
1355 } else {
1356 *flags &= ~BDRV_O_PROTOCOL;
1357 }
1358
1359 /* Translate cache options from flags into options */
1360 update_options_from_flags(*options, *flags);
1361
1362 /* Fetch the file name from the options QDict if necessary */
1363 if (protocol && filename) {
1364 if (!qdict_haskey(*options, "filename")) {
1365 qdict_put(*options, "filename", qstring_from_str(filename));
1366 parse_filename = true;
1367 } else {
1368 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1369 "the same time");
1370 return -EINVAL;
1371 }
1372 }
1373
1374 /* Find the right block driver */
1375 /* See cautionary note on accessing @options above */
1376 filename = qdict_get_try_str(*options, "filename");
1377
1378 if (!drvname && protocol) {
1379 if (filename) {
1380 drv = bdrv_find_protocol(filename, parse_filename, errp);
1381 if (!drv) {
1382 return -EINVAL;
1383 }
1384
1385 drvname = drv->format_name;
1386 qdict_put(*options, "driver", qstring_from_str(drvname));
1387 } else {
1388 error_setg(errp, "Must specify either driver or file");
1389 return -EINVAL;
1390 }
1391 }
1392
1393 assert(drv || !protocol);
1394
1395 /* Driver-specific filename parsing */
1396 if (drv && drv->bdrv_parse_filename && parse_filename) {
1397 drv->bdrv_parse_filename(filename, *options, &local_err);
1398 if (local_err) {
1399 error_propagate(errp, local_err);
1400 return -EINVAL;
1401 }
1402
1403 if (!drv->bdrv_needs_filename) {
1404 qdict_del(*options, "filename");
1405 }
1406 }
1407
1408 return 0;
1409 }
1410
1411 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1412 GSList *ignore_children, Error **errp);
1413 static void bdrv_child_abort_perm_update(BdrvChild *c);
1414 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1415
1416 /*
1417 * Check whether permissions on this node can be changed in a way that
1418 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1419 * permissions of all its parents. This involves checking whether all necessary
1420 * permission changes to child nodes can be performed.
1421 *
1422 * A call to this function must always be followed by a call to bdrv_set_perm()
1423 * or bdrv_abort_perm_update().
1424 */
1425 static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1426 uint64_t cumulative_shared_perms,
1427 GSList *ignore_children, Error **errp)
1428 {
1429 BlockDriver *drv = bs->drv;
1430 BdrvChild *c;
1431 int ret;
1432
1433 /* Write permissions never work with read-only images */
1434 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1435 bdrv_is_read_only(bs))
1436 {
1437 error_setg(errp, "Block node is read-only");
1438 return -EPERM;
1439 }
1440
1441 /* Check this node */
1442 if (!drv) {
1443 return 0;
1444 }
1445
1446 if (drv->bdrv_check_perm) {
1447 return drv->bdrv_check_perm(bs, cumulative_perms,
1448 cumulative_shared_perms, errp);
1449 }
1450
1451 /* Drivers that never have children can omit .bdrv_child_perm() */
1452 if (!drv->bdrv_child_perm) {
1453 assert(QLIST_EMPTY(&bs->children));
1454 return 0;
1455 }
1456
1457 /* Check all children */
1458 QLIST_FOREACH(c, &bs->children, next) {
1459 uint64_t cur_perm, cur_shared;
1460 drv->bdrv_child_perm(bs, c, c->role,
1461 cumulative_perms, cumulative_shared_perms,
1462 &cur_perm, &cur_shared);
1463 ret = bdrv_child_check_perm(c, cur_perm, cur_shared, ignore_children,
1464 errp);
1465 if (ret < 0) {
1466 return ret;
1467 }
1468 }
1469
1470 return 0;
1471 }
1472
1473 /*
1474 * Notifies drivers that after a previous bdrv_check_perm() call, the
1475 * permission update is not performed and any preparations made for it (e.g.
1476 * taken file locks) need to be undone.
1477 *
1478 * This function recursively notifies all child nodes.
1479 */
1480 static void bdrv_abort_perm_update(BlockDriverState *bs)
1481 {
1482 BlockDriver *drv = bs->drv;
1483 BdrvChild *c;
1484
1485 if (!drv) {
1486 return;
1487 }
1488
1489 if (drv->bdrv_abort_perm_update) {
1490 drv->bdrv_abort_perm_update(bs);
1491 }
1492
1493 QLIST_FOREACH(c, &bs->children, next) {
1494 bdrv_child_abort_perm_update(c);
1495 }
1496 }
1497
1498 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1499 uint64_t cumulative_shared_perms)
1500 {
1501 BlockDriver *drv = bs->drv;
1502 BdrvChild *c;
1503
1504 if (!drv) {
1505 return;
1506 }
1507
1508 /* Update this node */
1509 if (drv->bdrv_set_perm) {
1510 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1511 }
1512
1513 /* Drivers that never have children can omit .bdrv_child_perm() */
1514 if (!drv->bdrv_child_perm) {
1515 assert(QLIST_EMPTY(&bs->children));
1516 return;
1517 }
1518
1519 /* Update all children */
1520 QLIST_FOREACH(c, &bs->children, next) {
1521 uint64_t cur_perm, cur_shared;
1522 drv->bdrv_child_perm(bs, c, c->role,
1523 cumulative_perms, cumulative_shared_perms,
1524 &cur_perm, &cur_shared);
1525 bdrv_child_set_perm(c, cur_perm, cur_shared);
1526 }
1527 }
1528
1529 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1530 uint64_t *shared_perm)
1531 {
1532 BdrvChild *c;
1533 uint64_t cumulative_perms = 0;
1534 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1535
1536 QLIST_FOREACH(c, &bs->parents, next_parent) {
1537 cumulative_perms |= c->perm;
1538 cumulative_shared_perms &= c->shared_perm;
1539 }
1540
1541 *perm = cumulative_perms;
1542 *shared_perm = cumulative_shared_perms;
1543 }
1544
1545 static char *bdrv_child_user_desc(BdrvChild *c)
1546 {
1547 if (c->role->get_parent_desc) {
1548 return c->role->get_parent_desc(c);
1549 }
1550
1551 return g_strdup("another user");
1552 }
1553
1554 static char *bdrv_perm_names(uint64_t perm)
1555 {
1556 struct perm_name {
1557 uint64_t perm;
1558 const char *name;
1559 } permissions[] = {
1560 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1561 { BLK_PERM_WRITE, "write" },
1562 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1563 { BLK_PERM_RESIZE, "resize" },
1564 { BLK_PERM_GRAPH_MOD, "change children" },
1565 { 0, NULL }
1566 };
1567
1568 char *result = g_strdup("");
1569 struct perm_name *p;
1570
1571 for (p = permissions; p->name; p++) {
1572 if (perm & p->perm) {
1573 char *old = result;
1574 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1575 g_free(old);
1576 }
1577 }
1578
1579 return result;
1580 }
1581
1582 /*
1583 * Checks whether a new reference to @bs can be added if the new user requires
1584 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1585 * set, the BdrvChild objects in this list are ignored in the calculations;
1586 * this allows checking permission updates for an existing reference.
1587 *
1588 * Needs to be followed by a call to either bdrv_set_perm() or
1589 * bdrv_abort_perm_update(). */
1590 static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm,
1591 uint64_t new_shared_perm,
1592 GSList *ignore_children, Error **errp)
1593 {
1594 BdrvChild *c;
1595 uint64_t cumulative_perms = new_used_perm;
1596 uint64_t cumulative_shared_perms = new_shared_perm;
1597
1598 /* There is no reason why anyone couldn't tolerate write_unchanged */
1599 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1600
1601 QLIST_FOREACH(c, &bs->parents, next_parent) {
1602 if (g_slist_find(ignore_children, c)) {
1603 continue;
1604 }
1605
1606 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1607 char *user = bdrv_child_user_desc(c);
1608 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1609 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1610 "allow '%s' on %s",
1611 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1612 g_free(user);
1613 g_free(perm_names);
1614 return -EPERM;
1615 }
1616
1617 if ((c->perm & new_shared_perm) != c->perm) {
1618 char *user = bdrv_child_user_desc(c);
1619 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1620 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1621 "'%s' on %s",
1622 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1623 g_free(user);
1624 g_free(perm_names);
1625 return -EPERM;
1626 }
1627
1628 cumulative_perms |= c->perm;
1629 cumulative_shared_perms &= c->shared_perm;
1630 }
1631
1632 return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms,
1633 ignore_children, errp);
1634 }
1635
1636 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1637 * bdrv_child_abort_perm_update(). */
1638 static int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1639 GSList *ignore_children, Error **errp)
1640 {
1641 int ret;
1642
1643 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1644 ret = bdrv_check_update_perm(c->bs, perm, shared, ignore_children, errp);
1645 g_slist_free(ignore_children);
1646
1647 return ret;
1648 }
1649
1650 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1651 {
1652 uint64_t cumulative_perms, cumulative_shared_perms;
1653
1654 c->perm = perm;
1655 c->shared_perm = shared;
1656
1657 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1658 &cumulative_shared_perms);
1659 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1660 }
1661
1662 static void bdrv_child_abort_perm_update(BdrvChild *c)
1663 {
1664 bdrv_abort_perm_update(c->bs);
1665 }
1666
1667 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1668 Error **errp)
1669 {
1670 int ret;
1671
1672 ret = bdrv_child_check_perm(c, perm, shared, NULL, errp);
1673 if (ret < 0) {
1674 bdrv_child_abort_perm_update(c);
1675 return ret;
1676 }
1677
1678 bdrv_child_set_perm(c, perm, shared);
1679
1680 return 0;
1681 }
1682
1683 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1684 | BLK_PERM_WRITE \
1685 | BLK_PERM_WRITE_UNCHANGED \
1686 | BLK_PERM_RESIZE)
1687 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1688
1689 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1690 const BdrvChildRole *role,
1691 uint64_t perm, uint64_t shared,
1692 uint64_t *nperm, uint64_t *nshared)
1693 {
1694 if (c == NULL) {
1695 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1696 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1697 return;
1698 }
1699
1700 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1701 (c->perm & DEFAULT_PERM_UNCHANGED);
1702 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1703 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1704 }
1705
1706 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1707 const BdrvChildRole *role,
1708 uint64_t perm, uint64_t shared,
1709 uint64_t *nperm, uint64_t *nshared)
1710 {
1711 bool backing = (role == &child_backing);
1712 assert(role == &child_backing || role == &child_file);
1713
1714 if (!backing) {
1715 /* Apart from the modifications below, the same permissions are
1716 * forwarded and left alone as for filters */
1717 bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared);
1718
1719 /* Format drivers may touch metadata even if the guest doesn't write */
1720 if (!bdrv_is_read_only(bs)) {
1721 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1722 }
1723
1724 /* bs->file always needs to be consistent because of the metadata. We
1725 * can never allow other users to resize or write to it. */
1726 perm |= BLK_PERM_CONSISTENT_READ;
1727 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1728 } else {
1729 /* We want consistent read from backing files if the parent needs it.
1730 * No other operations are performed on backing files. */
1731 perm &= BLK_PERM_CONSISTENT_READ;
1732
1733 /* If the parent can deal with changing data, we're okay with a
1734 * writable and resizable backing file. */
1735 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1736 if (shared & BLK_PERM_WRITE) {
1737 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1738 } else {
1739 shared = 0;
1740 }
1741
1742 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1743 BLK_PERM_WRITE_UNCHANGED;
1744 }
1745
1746 *nperm = perm;
1747 *nshared = shared;
1748 }
1749
1750 static void bdrv_replace_child_noperm(BdrvChild *child,
1751 BlockDriverState *new_bs)
1752 {
1753 BlockDriverState *old_bs = child->bs;
1754
1755 if (old_bs && new_bs) {
1756 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1757 }
1758 if (old_bs) {
1759 if (old_bs->quiesce_counter && child->role->drained_end) {
1760 child->role->drained_end(child);
1761 }
1762 if (child->role->detach) {
1763 child->role->detach(child);
1764 }
1765 QLIST_REMOVE(child, next_parent);
1766 }
1767
1768 child->bs = new_bs;
1769
1770 if (new_bs) {
1771 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1772 if (new_bs->quiesce_counter && child->role->drained_begin) {
1773 child->role->drained_begin(child);
1774 }
1775
1776 if (child->role->attach) {
1777 child->role->attach(child);
1778 }
1779 }
1780 }
1781
1782 /*
1783 * Updates @child to change its reference to point to @new_bs, including
1784 * checking and applying the necessary permisson updates both to the old node
1785 * and to @new_bs.
1786 *
1787 * NULL is passed as @new_bs for removing the reference before freeing @child.
1788 *
1789 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1790 * function uses bdrv_set_perm() to update the permissions according to the new
1791 * reference that @new_bs gets.
1792 */
1793 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1794 {
1795 BlockDriverState *old_bs = child->bs;
1796 uint64_t perm, shared_perm;
1797
1798 if (old_bs) {
1799 /* Update permissions for old node. This is guaranteed to succeed
1800 * because we're just taking a parent away, so we're loosening
1801 * restrictions. */
1802 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1803 bdrv_check_perm(old_bs, perm, shared_perm, NULL, &error_abort);
1804 bdrv_set_perm(old_bs, perm, shared_perm);
1805 }
1806
1807 bdrv_replace_child_noperm(child, new_bs);
1808
1809 if (new_bs) {
1810 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
1811 bdrv_set_perm(new_bs, perm, shared_perm);
1812 }
1813 }
1814
1815 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1816 const char *child_name,
1817 const BdrvChildRole *child_role,
1818 uint64_t perm, uint64_t shared_perm,
1819 void *opaque, Error **errp)
1820 {
1821 BdrvChild *child;
1822 int ret;
1823
1824 ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp);
1825 if (ret < 0) {
1826 bdrv_abort_perm_update(child_bs);
1827 return NULL;
1828 }
1829
1830 child = g_new(BdrvChild, 1);
1831 *child = (BdrvChild) {
1832 .bs = NULL,
1833 .name = g_strdup(child_name),
1834 .role = child_role,
1835 .perm = perm,
1836 .shared_perm = shared_perm,
1837 .opaque = opaque,
1838 };
1839
1840 /* This performs the matching bdrv_set_perm() for the above check. */
1841 bdrv_replace_child(child, child_bs);
1842
1843 return child;
1844 }
1845
1846 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1847 BlockDriverState *child_bs,
1848 const char *child_name,
1849 const BdrvChildRole *child_role,
1850 Error **errp)
1851 {
1852 BdrvChild *child;
1853 uint64_t perm, shared_perm;
1854
1855 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
1856
1857 assert(parent_bs->drv);
1858 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
1859 parent_bs->drv->bdrv_child_perm(parent_bs, NULL, child_role,
1860 perm, shared_perm, &perm, &shared_perm);
1861
1862 child = bdrv_root_attach_child(child_bs, child_name, child_role,
1863 perm, shared_perm, parent_bs, errp);
1864 if (child == NULL) {
1865 return NULL;
1866 }
1867
1868 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1869 return child;
1870 }
1871
1872 static void bdrv_detach_child(BdrvChild *child)
1873 {
1874 if (child->next.le_prev) {
1875 QLIST_REMOVE(child, next);
1876 child->next.le_prev = NULL;
1877 }
1878
1879 bdrv_replace_child(child, NULL);
1880
1881 g_free(child->name);
1882 g_free(child);
1883 }
1884
1885 void bdrv_root_unref_child(BdrvChild *child)
1886 {
1887 BlockDriverState *child_bs;
1888
1889 child_bs = child->bs;
1890 bdrv_detach_child(child);
1891 bdrv_unref(child_bs);
1892 }
1893
1894 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1895 {
1896 if (child == NULL) {
1897 return;
1898 }
1899
1900 if (child->bs->inherits_from == parent) {
1901 BdrvChild *c;
1902
1903 /* Remove inherits_from only when the last reference between parent and
1904 * child->bs goes away. */
1905 QLIST_FOREACH(c, &parent->children, next) {
1906 if (c != child && c->bs == child->bs) {
1907 break;
1908 }
1909 }
1910 if (c == NULL) {
1911 child->bs->inherits_from = NULL;
1912 }
1913 }
1914
1915 bdrv_root_unref_child(child);
1916 }
1917
1918
1919 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1920 {
1921 BdrvChild *c;
1922 QLIST_FOREACH(c, &bs->parents, next_parent) {
1923 if (c->role->change_media) {
1924 c->role->change_media(c, load);
1925 }
1926 }
1927 }
1928
1929 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1930 {
1931 BdrvChild *c;
1932 QLIST_FOREACH(c, &bs->parents, next_parent) {
1933 if (c->role->resize) {
1934 c->role->resize(c);
1935 }
1936 }
1937 }
1938
1939 /*
1940 * Sets the backing file link of a BDS. A new reference is created; callers
1941 * which don't need their own reference any more must call bdrv_unref().
1942 */
1943 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
1944 Error **errp)
1945 {
1946 if (backing_hd) {
1947 bdrv_ref(backing_hd);
1948 }
1949
1950 if (bs->backing) {
1951 bdrv_unref_child(bs, bs->backing);
1952 }
1953
1954 if (!backing_hd) {
1955 bs->backing = NULL;
1956 goto out;
1957 }
1958
1959 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
1960 errp);
1961 if (!bs->backing) {
1962 bdrv_unref(backing_hd);
1963 }
1964
1965 bdrv_refresh_filename(bs);
1966
1967 out:
1968 bdrv_refresh_limits(bs, NULL);
1969 }
1970
1971 /*
1972 * Opens the backing file for a BlockDriverState if not yet open
1973 *
1974 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1975 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1976 * itself, all options starting with "${bdref_key}." are considered part of the
1977 * BlockdevRef.
1978 *
1979 * TODO Can this be unified with bdrv_open_image()?
1980 */
1981 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1982 const char *bdref_key, Error **errp)
1983 {
1984 char *backing_filename = g_malloc0(PATH_MAX);
1985 char *bdref_key_dot;
1986 const char *reference = NULL;
1987 int ret = 0;
1988 BlockDriverState *backing_hd;
1989 QDict *options;
1990 QDict *tmp_parent_options = NULL;
1991 Error *local_err = NULL;
1992
1993 if (bs->backing != NULL) {
1994 goto free_exit;
1995 }
1996
1997 /* NULL means an empty set of options */
1998 if (parent_options == NULL) {
1999 tmp_parent_options = qdict_new();
2000 parent_options = tmp_parent_options;
2001 }
2002
2003 bs->open_flags &= ~BDRV_O_NO_BACKING;
2004
2005 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2006 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2007 g_free(bdref_key_dot);
2008
2009 /*
2010 * Caution: while qdict_get_try_str() is fine, getting non-string
2011 * types would require more care. When @parent_options come from
2012 * -blockdev or blockdev_add, its members are typed according to
2013 * the QAPI schema, but when they come from -drive, they're all
2014 * QString.
2015 */
2016 reference = qdict_get_try_str(parent_options, bdref_key);
2017 if (reference || qdict_haskey(options, "file.filename")) {
2018 backing_filename[0] = '\0';
2019 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2020 QDECREF(options);
2021 goto free_exit;
2022 } else {
2023 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2024 &local_err);
2025 if (local_err) {
2026 ret = -EINVAL;
2027 error_propagate(errp, local_err);
2028 QDECREF(options);
2029 goto free_exit;
2030 }
2031 }
2032
2033 if (!bs->drv || !bs->drv->supports_backing) {
2034 ret = -EINVAL;
2035 error_setg(errp, "Driver doesn't support backing files");
2036 QDECREF(options);
2037 goto free_exit;
2038 }
2039
2040 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2041 qdict_put(options, "driver", qstring_from_str(bs->backing_format));
2042 }
2043
2044 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2045 reference, options, 0, bs, &child_backing,
2046 errp);
2047 if (!backing_hd) {
2048 bs->open_flags |= BDRV_O_NO_BACKING;
2049 error_prepend(errp, "Could not open backing file: ");
2050 ret = -EINVAL;
2051 goto free_exit;
2052 }
2053
2054 /* Hook up the backing file link; drop our reference, bs owns the
2055 * backing_hd reference now */
2056 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2057 bdrv_unref(backing_hd);
2058 if (local_err) {
2059 error_propagate(errp, local_err);
2060 ret = -EINVAL;
2061 goto free_exit;
2062 }
2063
2064 qdict_del(parent_options, bdref_key);
2065
2066 free_exit:
2067 g_free(backing_filename);
2068 QDECREF(tmp_parent_options);
2069 return ret;
2070 }
2071
2072 static BlockDriverState *
2073 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2074 BlockDriverState *parent, const BdrvChildRole *child_role,
2075 bool allow_none, Error **errp)
2076 {
2077 BlockDriverState *bs = NULL;
2078 QDict *image_options;
2079 char *bdref_key_dot;
2080 const char *reference;
2081
2082 assert(child_role != NULL);
2083
2084 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2085 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2086 g_free(bdref_key_dot);
2087
2088 /*
2089 * Caution: while qdict_get_try_str() is fine, getting non-string
2090 * types would require more care. When @options come from
2091 * -blockdev or blockdev_add, its members are typed according to
2092 * the QAPI schema, but when they come from -drive, they're all
2093 * QString.
2094 */
2095 reference = qdict_get_try_str(options, bdref_key);
2096 if (!filename && !reference && !qdict_size(image_options)) {
2097 if (!allow_none) {
2098 error_setg(errp, "A block device must be specified for \"%s\"",
2099 bdref_key);
2100 }
2101 QDECREF(image_options);
2102 goto done;
2103 }
2104
2105 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2106 parent, child_role, errp);
2107 if (!bs) {
2108 goto done;
2109 }
2110
2111 done:
2112 qdict_del(options, bdref_key);
2113 return bs;
2114 }
2115
2116 /*
2117 * Opens a disk image whose options are given as BlockdevRef in another block
2118 * device's options.
2119 *
2120 * If allow_none is true, no image will be opened if filename is false and no
2121 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2122 *
2123 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2124 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2125 * itself, all options starting with "${bdref_key}." are considered part of the
2126 * BlockdevRef.
2127 *
2128 * The BlockdevRef will be removed from the options QDict.
2129 */
2130 BdrvChild *bdrv_open_child(const char *filename,
2131 QDict *options, const char *bdref_key,
2132 BlockDriverState *parent,
2133 const BdrvChildRole *child_role,
2134 bool allow_none, Error **errp)
2135 {
2136 BdrvChild *c;
2137 BlockDriverState *bs;
2138
2139 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2140 allow_none, errp);
2141 if (bs == NULL) {
2142 return NULL;
2143 }
2144
2145 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2146 if (!c) {
2147 bdrv_unref(bs);
2148 return NULL;
2149 }
2150
2151 return c;
2152 }
2153
2154 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2155 int flags,
2156 QDict *snapshot_options,
2157 Error **errp)
2158 {
2159 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2160 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2161 int64_t total_size;
2162 QemuOpts *opts = NULL;
2163 BlockDriverState *bs_snapshot;
2164 Error *local_err = NULL;
2165 int ret;
2166
2167 /* if snapshot, we create a temporary backing file and open it
2168 instead of opening 'filename' directly */
2169
2170 /* Get the required size from the image */
2171 total_size = bdrv_getlength(bs);
2172 if (total_size < 0) {
2173 error_setg_errno(errp, -total_size, "Could not get image size");
2174 goto out;
2175 }
2176
2177 /* Create the temporary image */
2178 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2179 if (ret < 0) {
2180 error_setg_errno(errp, -ret, "Could not get temporary filename");
2181 goto out;
2182 }
2183
2184 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2185 &error_abort);
2186 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2187 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2188 qemu_opts_del(opts);
2189 if (ret < 0) {
2190 error_prepend(errp, "Could not create temporary overlay '%s': ",
2191 tmp_filename);
2192 goto out;
2193 }
2194
2195 /* Prepare options QDict for the temporary file */
2196 qdict_put(snapshot_options, "file.driver",
2197 qstring_from_str("file"));
2198 qdict_put(snapshot_options, "file.filename",
2199 qstring_from_str(tmp_filename));
2200 qdict_put(snapshot_options, "driver",
2201 qstring_from_str("qcow2"));
2202
2203 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2204 snapshot_options = NULL;
2205 if (!bs_snapshot) {
2206 ret = -EINVAL;
2207 goto out;
2208 }
2209
2210 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
2211 * call bdrv_unref() on it), so in order to be able to return one, we have
2212 * to increase bs_snapshot's refcount here */
2213 bdrv_ref(bs_snapshot);
2214 bdrv_append(bs_snapshot, bs, &local_err);
2215 if (local_err) {
2216 error_propagate(errp, local_err);
2217 ret = -EINVAL;
2218 goto out;
2219 }
2220
2221 g_free(tmp_filename);
2222 return bs_snapshot;
2223
2224 out:
2225 QDECREF(snapshot_options);
2226 g_free(tmp_filename);
2227 return NULL;
2228 }
2229
2230 /*
2231 * Opens a disk image (raw, qcow2, vmdk, ...)
2232 *
2233 * options is a QDict of options to pass to the block drivers, or NULL for an
2234 * empty set of options. The reference to the QDict belongs to the block layer
2235 * after the call (even on failure), so if the caller intends to reuse the
2236 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2237 *
2238 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2239 * If it is not NULL, the referenced BDS will be reused.
2240 *
2241 * The reference parameter may be used to specify an existing block device which
2242 * should be opened. If specified, neither options nor a filename may be given,
2243 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2244 */
2245 static BlockDriverState *bdrv_open_inherit(const char *filename,
2246 const char *reference,
2247 QDict *options, int flags,
2248 BlockDriverState *parent,
2249 const BdrvChildRole *child_role,
2250 Error **errp)
2251 {
2252 int ret;
2253 BlockBackend *file = NULL;
2254 BlockDriverState *bs;
2255 BlockDriver *drv = NULL;
2256 const char *drvname;
2257 const char *backing;
2258 Error *local_err = NULL;
2259 QDict *snapshot_options = NULL;
2260 int snapshot_flags = 0;
2261
2262 assert(!child_role || !flags);
2263 assert(!child_role == !parent);
2264
2265 if (reference) {
2266 bool options_non_empty = options ? qdict_size(options) : false;
2267 QDECREF(options);
2268
2269 if (filename || options_non_empty) {
2270 error_setg(errp, "Cannot reference an existing block device with "
2271 "additional options or a new filename");
2272 return NULL;
2273 }
2274
2275 bs = bdrv_lookup_bs(reference, reference, errp);
2276 if (!bs) {
2277 return NULL;
2278 }
2279
2280 bdrv_ref(bs);
2281 return bs;
2282 }
2283
2284 bs = bdrv_new();
2285
2286 /* NULL means an empty set of options */
2287 if (options == NULL) {
2288 options = qdict_new();
2289 }
2290
2291 /* json: syntax counts as explicit options, as if in the QDict */
2292 parse_json_protocol(options, &filename, &local_err);
2293 if (local_err) {
2294 goto fail;
2295 }
2296
2297 bs->explicit_options = qdict_clone_shallow(options);
2298
2299 if (child_role) {
2300 bs->inherits_from = parent;
2301 child_role->inherit_options(&flags, options,
2302 parent->open_flags, parent->options);
2303 }
2304
2305 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2306 if (local_err) {
2307 goto fail;
2308 }
2309
2310 /*
2311 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2312 * Caution: getting a boolean member of @options requires care.
2313 * When @options come from -blockdev or blockdev_add, members are
2314 * typed according to the QAPI schema, but when they come from
2315 * -drive, they're all QString.
2316 */
2317 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2318 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2319 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2320 } else {
2321 flags &= ~BDRV_O_RDWR;
2322 }
2323
2324 if (flags & BDRV_O_SNAPSHOT) {
2325 snapshot_options = qdict_new();
2326 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2327 flags, options);
2328 /* Let bdrv_backing_options() override "read-only" */
2329 qdict_del(options, BDRV_OPT_READ_ONLY);
2330 bdrv_backing_options(&flags, options, flags, options);
2331 }
2332
2333 bs->open_flags = flags;
2334 bs->options = options;
2335 options = qdict_clone_shallow(options);
2336
2337 /* Find the right image format driver */
2338 /* See cautionary note on accessing @options above */
2339 drvname = qdict_get_try_str(options, "driver");
2340 if (drvname) {
2341 drv = bdrv_find_format(drvname);
2342 if (!drv) {
2343 error_setg(errp, "Unknown driver: '%s'", drvname);
2344 goto fail;
2345 }
2346 }
2347
2348 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2349
2350 /* See cautionary note on accessing @options above */
2351 backing = qdict_get_try_str(options, "backing");
2352 if (backing && *backing == '\0') {
2353 flags |= BDRV_O_NO_BACKING;
2354 qdict_del(options, "backing");
2355 }
2356
2357 /* Open image file without format layer. This BlockBackend is only used for
2358 * probing, the block drivers will do their own bdrv_open_child() for the
2359 * same BDS, which is why we put the node name back into options. */
2360 if ((flags & BDRV_O_PROTOCOL) == 0) {
2361 BlockDriverState *file_bs;
2362
2363 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2364 &child_file, true, &local_err);
2365 if (local_err) {
2366 goto fail;
2367 }
2368 if (file_bs != NULL) {
2369 file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
2370 blk_insert_bs(file, file_bs, &local_err);
2371 bdrv_unref(file_bs);
2372 if (local_err) {
2373 goto fail;
2374 }
2375
2376 qdict_put(options, "file",
2377 qstring_from_str(bdrv_get_node_name(file_bs)));
2378 }
2379 }
2380
2381 /* Image format probing */
2382 bs->probed = !drv;
2383 if (!drv && file) {
2384 ret = find_image_format(file, filename, &drv, &local_err);
2385 if (ret < 0) {
2386 goto fail;
2387 }
2388 /*
2389 * This option update would logically belong in bdrv_fill_options(),
2390 * but we first need to open bs->file for the probing to work, while
2391 * opening bs->file already requires the (mostly) final set of options
2392 * so that cache mode etc. can be inherited.
2393 *
2394 * Adding the driver later is somewhat ugly, but it's not an option
2395 * that would ever be inherited, so it's correct. We just need to make
2396 * sure to update both bs->options (which has the full effective
2397 * options for bs) and options (which has file.* already removed).
2398 */
2399 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
2400 qdict_put(options, "driver", qstring_from_str(drv->format_name));
2401 } else if (!drv) {
2402 error_setg(errp, "Must specify either driver or file");
2403 goto fail;
2404 }
2405
2406 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2407 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2408 /* file must be NULL if a protocol BDS is about to be created
2409 * (the inverse results in an error message from bdrv_open_common()) */
2410 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2411
2412 /* Open the image */
2413 ret = bdrv_open_common(bs, file, options, &local_err);
2414 if (ret < 0) {
2415 goto fail;
2416 }
2417
2418 if (file) {
2419 blk_unref(file);
2420 file = NULL;
2421 }
2422
2423 /* If there is a backing file, use it */
2424 if ((flags & BDRV_O_NO_BACKING) == 0) {
2425 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2426 if (ret < 0) {
2427 goto close_and_fail;
2428 }
2429 }
2430
2431 bdrv_refresh_filename(bs);
2432
2433 /* Check if any unknown options were used */
2434 if (qdict_size(options) != 0) {
2435 const QDictEntry *entry = qdict_first(options);
2436 if (flags & BDRV_O_PROTOCOL) {
2437 error_setg(errp, "Block protocol '%s' doesn't support the option "
2438 "'%s'", drv->format_name, entry->key);
2439 } else {
2440 error_setg(errp,
2441 "Block format '%s' does not support the option '%s'",
2442 drv->format_name, entry->key);
2443 }
2444
2445 goto close_and_fail;
2446 }
2447
2448 if (!bdrv_key_required(bs)) {
2449 bdrv_parent_cb_change_media(bs, true);
2450 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
2451 && !runstate_check(RUN_STATE_INMIGRATE)
2452 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
2453 error_setg(errp,
2454 "Guest must be stopped for opening of encrypted image");
2455 goto close_and_fail;
2456 }
2457
2458 QDECREF(options);
2459
2460 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2461 * temporary snapshot afterwards. */
2462 if (snapshot_flags) {
2463 BlockDriverState *snapshot_bs;
2464 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2465 snapshot_options, &local_err);
2466 snapshot_options = NULL;
2467 if (local_err) {
2468 goto close_and_fail;
2469 }
2470 /* We are not going to return bs but the overlay on top of it
2471 * (snapshot_bs); thus, we have to drop the strong reference to bs
2472 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2473 * though, because the overlay still has a reference to it. */
2474 bdrv_unref(bs);
2475 bs = snapshot_bs;
2476 }
2477
2478 return bs;
2479
2480 fail:
2481 blk_unref(file);
2482 if (bs->file != NULL) {
2483 bdrv_unref_child(bs, bs->file);
2484 }
2485 QDECREF(snapshot_options);
2486 QDECREF(bs->explicit_options);
2487 QDECREF(bs->options);
2488 QDECREF(options);
2489 bs->options = NULL;
2490 bdrv_unref(bs);
2491 error_propagate(errp, local_err);
2492 return NULL;
2493
2494 close_and_fail:
2495 bdrv_unref(bs);
2496 QDECREF(snapshot_options);
2497 QDECREF(options);
2498 error_propagate(errp, local_err);
2499 return NULL;
2500 }
2501
2502 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2503 QDict *options, int flags, Error **errp)
2504 {
2505 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2506 NULL, errp);
2507 }
2508
2509 typedef struct BlockReopenQueueEntry {
2510 bool prepared;
2511 BDRVReopenState state;
2512 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2513 } BlockReopenQueueEntry;
2514
2515 /*
2516 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2517 * reopen of multiple devices.
2518 *
2519 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2520 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2521 * be created and initialized. This newly created BlockReopenQueue should be
2522 * passed back in for subsequent calls that are intended to be of the same
2523 * atomic 'set'.
2524 *
2525 * bs is the BlockDriverState to add to the reopen queue.
2526 *
2527 * options contains the changed options for the associated bs
2528 * (the BlockReopenQueue takes ownership)
2529 *
2530 * flags contains the open flags for the associated bs
2531 *
2532 * returns a pointer to bs_queue, which is either the newly allocated
2533 * bs_queue, or the existing bs_queue being used.
2534 *
2535 */
2536 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2537 BlockDriverState *bs,
2538 QDict *options,
2539 int flags,
2540 const BdrvChildRole *role,
2541 QDict *parent_options,
2542 int parent_flags)
2543 {
2544 assert(bs != NULL);
2545
2546 BlockReopenQueueEntry *bs_entry;
2547 BdrvChild *child;
2548 QDict *old_options, *explicit_options;
2549
2550 if (bs_queue == NULL) {
2551 bs_queue = g_new0(BlockReopenQueue, 1);
2552 QSIMPLEQ_INIT(bs_queue);
2553 }
2554
2555 if (!options) {
2556 options = qdict_new();
2557 }
2558
2559 /* Check if this BlockDriverState is already in the queue */
2560 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2561 if (bs == bs_entry->state.bs) {
2562 break;
2563 }
2564 }
2565
2566 /*
2567 * Precedence of options:
2568 * 1. Explicitly passed in options (highest)
2569 * 2. Set in flags (only for top level)
2570 * 3. Retained from explicitly set options of bs
2571 * 4. Inherited from parent node
2572 * 5. Retained from effective options of bs
2573 */
2574
2575 if (!parent_options) {
2576 /*
2577 * Any setting represented by flags is always updated. If the
2578 * corresponding QDict option is set, it takes precedence. Otherwise
2579 * the flag is translated into a QDict option. The old setting of bs is
2580 * not considered.
2581 */
2582 update_options_from_flags(options, flags);
2583 }
2584
2585 /* Old explicitly set values (don't overwrite by inherited value) */
2586 if (bs_entry) {
2587 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2588 } else {
2589 old_options = qdict_clone_shallow(bs->explicit_options);
2590 }
2591 bdrv_join_options(bs, options, old_options);
2592 QDECREF(old_options);
2593
2594 explicit_options = qdict_clone_shallow(options);
2595
2596 /* Inherit from parent node */
2597 if (parent_options) {
2598 assert(!flags);
2599 role->inherit_options(&flags, options, parent_flags, parent_options);
2600 }
2601
2602 /* Old values are used for options that aren't set yet */
2603 old_options = qdict_clone_shallow(bs->options);
2604 bdrv_join_options(bs, options, old_options);
2605 QDECREF(old_options);
2606
2607 /* bdrv_open() masks this flag out */
2608 flags &= ~BDRV_O_PROTOCOL;
2609
2610 QLIST_FOREACH(child, &bs->children, next) {
2611 QDict *new_child_options;
2612 char *child_key_dot;
2613
2614 /* reopen can only change the options of block devices that were
2615 * implicitly created and inherited options. For other (referenced)
2616 * block devices, a syntax like "backing.foo" results in an error. */
2617 if (child->bs->inherits_from != bs) {
2618 continue;
2619 }
2620
2621 child_key_dot = g_strdup_printf("%s.", child->name);
2622 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2623 g_free(child_key_dot);
2624
2625 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2626 child->role, options, flags);
2627 }
2628
2629 if (!bs_entry) {
2630 bs_entry = g_new0(BlockReopenQueueEntry, 1);
2631 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2632 } else {
2633 QDECREF(bs_entry->state.options);
2634 QDECREF(bs_entry->state.explicit_options);
2635 }
2636
2637 bs_entry->state.bs = bs;
2638 bs_entry->state.options = options;
2639 bs_entry->state.explicit_options = explicit_options;
2640 bs_entry->state.flags = flags;
2641
2642 return bs_queue;
2643 }
2644
2645 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2646 BlockDriverState *bs,
2647 QDict *options, int flags)
2648 {
2649 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2650 NULL, NULL, 0);
2651 }
2652
2653 /*
2654 * Reopen multiple BlockDriverStates atomically & transactionally.
2655 *
2656 * The queue passed in (bs_queue) must have been built up previous
2657 * via bdrv_reopen_queue().
2658 *
2659 * Reopens all BDS specified in the queue, with the appropriate
2660 * flags. All devices are prepared for reopen, and failure of any
2661 * device will cause all device changes to be abandonded, and intermediate
2662 * data cleaned up.
2663 *
2664 * If all devices prepare successfully, then the changes are committed
2665 * to all devices.
2666 *
2667 */
2668 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2669 {
2670 int ret = -1;
2671 BlockReopenQueueEntry *bs_entry, *next;
2672 Error *local_err = NULL;
2673
2674 assert(bs_queue != NULL);
2675
2676 aio_context_release(ctx);
2677 bdrv_drain_all_begin();
2678 aio_context_acquire(ctx);
2679
2680 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2681 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2682 error_propagate(errp, local_err);
2683 goto cleanup;
2684 }
2685 bs_entry->prepared = true;
2686 }
2687
2688 /* If we reach this point, we have success and just need to apply the
2689 * changes
2690 */
2691 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2692 bdrv_reopen_commit(&bs_entry->state);
2693 }
2694
2695 ret = 0;
2696
2697 cleanup:
2698 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2699 if (ret && bs_entry->prepared) {
2700 bdrv_reopen_abort(&bs_entry->state);
2701 } else if (ret) {
2702 QDECREF(bs_entry->state.explicit_options);
2703 }
2704 QDECREF(bs_entry->state.options);
2705 g_free(bs_entry);
2706 }
2707 g_free(bs_queue);
2708
2709 bdrv_drain_all_end();
2710
2711 return ret;
2712 }
2713
2714
2715 /* Reopen a single BlockDriverState with the specified flags. */
2716 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2717 {
2718 int ret = -1;
2719 Error *local_err = NULL;
2720 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2721
2722 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2723 if (local_err != NULL) {
2724 error_propagate(errp, local_err);
2725 }
2726 return ret;
2727 }
2728
2729
2730 /*
2731 * Prepares a BlockDriverState for reopen. All changes are staged in the
2732 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2733 * the block driver layer .bdrv_reopen_prepare()
2734 *
2735 * bs is the BlockDriverState to reopen
2736 * flags are the new open flags
2737 * queue is the reopen queue
2738 *
2739 * Returns 0 on success, non-zero on error. On error errp will be set
2740 * as well.
2741 *
2742 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2743 * It is the responsibility of the caller to then call the abort() or
2744 * commit() for any other BDS that have been left in a prepare() state
2745 *
2746 */
2747 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2748 Error **errp)
2749 {
2750 int ret = -1;
2751 Error *local_err = NULL;
2752 BlockDriver *drv;
2753 QemuOpts *opts;
2754 const char *value;
2755
2756 assert(reopen_state != NULL);
2757 assert(reopen_state->bs->drv != NULL);
2758 drv = reopen_state->bs->drv;
2759
2760 /* Process generic block layer options */
2761 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2762 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2763 if (local_err) {
2764 error_propagate(errp, local_err);
2765 ret = -EINVAL;
2766 goto error;
2767 }
2768
2769 update_flags_from_options(&reopen_state->flags, opts);
2770
2771 /* node-name and driver must be unchanged. Put them back into the QDict, so
2772 * that they are checked at the end of this function. */
2773 value = qemu_opt_get(opts, "node-name");
2774 if (value) {
2775 qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2776 }
2777
2778 value = qemu_opt_get(opts, "driver");
2779 if (value) {
2780 qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2781 }
2782
2783 /* if we are to stay read-only, do not allow permission change
2784 * to r/w */
2785 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2786 reopen_state->flags & BDRV_O_RDWR) {
2787 error_setg(errp, "Node '%s' is read only",
2788 bdrv_get_device_or_node_name(reopen_state->bs));
2789 goto error;
2790 }
2791
2792
2793 ret = bdrv_flush(reopen_state->bs);
2794 if (ret) {
2795 error_setg_errno(errp, -ret, "Error flushing drive");
2796 goto error;
2797 }
2798
2799 if (drv->bdrv_reopen_prepare) {
2800 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2801 if (ret) {
2802 if (local_err != NULL) {
2803 error_propagate(errp, local_err);
2804 } else {
2805 error_setg(errp, "failed while preparing to reopen image '%s'",
2806 reopen_state->bs->filename);
2807 }
2808 goto error;
2809 }
2810 } else {
2811 /* It is currently mandatory to have a bdrv_reopen_prepare()
2812 * handler for each supported drv. */
2813 error_setg(errp, "Block format '%s' used by node '%s' "
2814 "does not support reopening files", drv->format_name,
2815 bdrv_get_device_or_node_name(reopen_state->bs));
2816 ret = -1;
2817 goto error;
2818 }
2819
2820 /* Options that are not handled are only okay if they are unchanged
2821 * compared to the old state. It is expected that some options are only
2822 * used for the initial open, but not reopen (e.g. filename) */
2823 if (qdict_size(reopen_state->options)) {
2824 const QDictEntry *entry = qdict_first(reopen_state->options);
2825
2826 do {
2827 QString *new_obj = qobject_to_qstring(entry->value);
2828 const char *new = qstring_get_str(new_obj);
2829 /*
2830 * Caution: while qdict_get_try_str() is fine, getting
2831 * non-string types would require more care. When
2832 * bs->options come from -blockdev or blockdev_add, its
2833 * members are typed according to the QAPI schema, but
2834 * when they come from -drive, they're all QString.
2835 */
2836 const char *old = qdict_get_try_str(reopen_state->bs->options,
2837 entry->key);
2838
2839 if (!old || strcmp(new, old)) {
2840 error_setg(errp, "Cannot change the option '%s'", entry->key);
2841 ret = -EINVAL;
2842 goto error;
2843 }
2844 } while ((entry = qdict_next(reopen_state->options, entry)));
2845 }
2846
2847 ret = 0;
2848
2849 error:
2850 qemu_opts_del(opts);
2851 return ret;
2852 }
2853
2854 /*
2855 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2856 * makes them final by swapping the staging BlockDriverState contents into
2857 * the active BlockDriverState contents.
2858 */
2859 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2860 {
2861 BlockDriver *drv;
2862
2863 assert(reopen_state != NULL);
2864 drv = reopen_state->bs->drv;
2865 assert(drv != NULL);
2866
2867 /* If there are any driver level actions to take */
2868 if (drv->bdrv_reopen_commit) {
2869 drv->bdrv_reopen_commit(reopen_state);
2870 }
2871
2872 /* set BDS specific flags now */
2873 QDECREF(reopen_state->bs->explicit_options);
2874
2875 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2876 reopen_state->bs->open_flags = reopen_state->flags;
2877 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2878
2879 bdrv_refresh_limits(reopen_state->bs, NULL);
2880 }
2881
2882 /*
2883 * Abort the reopen, and delete and free the staged changes in
2884 * reopen_state
2885 */
2886 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2887 {
2888 BlockDriver *drv;
2889
2890 assert(reopen_state != NULL);
2891 drv = reopen_state->bs->drv;
2892 assert(drv != NULL);
2893
2894 if (drv->bdrv_reopen_abort) {
2895 drv->bdrv_reopen_abort(reopen_state);
2896 }
2897
2898 QDECREF(reopen_state->explicit_options);
2899 }
2900
2901
2902 static void bdrv_close(BlockDriverState *bs)
2903 {
2904 BdrvAioNotifier *ban, *ban_next;
2905
2906 assert(!bs->job);
2907 assert(!bs->refcnt);
2908
2909 bdrv_drained_begin(bs); /* complete I/O */
2910 bdrv_flush(bs);
2911 bdrv_drain(bs); /* in case flush left pending I/O */
2912
2913 bdrv_release_named_dirty_bitmaps(bs);
2914 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2915
2916 if (bs->drv) {
2917 BdrvChild *child, *next;
2918
2919 bs->drv->bdrv_close(bs);
2920 bs->drv = NULL;
2921
2922 bdrv_set_backing_hd(bs, NULL, &error_abort);
2923
2924 if (bs->file != NULL) {
2925 bdrv_unref_child(bs, bs->file);
2926 bs->file = NULL;
2927 }
2928
2929 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2930 /* TODO Remove bdrv_unref() from drivers' close function and use
2931 * bdrv_unref_child() here */
2932 if (child->bs->inherits_from == bs) {
2933 child->bs->inherits_from = NULL;
2934 }
2935 bdrv_detach_child(child);
2936 }
2937
2938 g_free(bs->opaque);
2939 bs->opaque = NULL;
2940 bs->copy_on_read = 0;
2941 bs->backing_file[0] = '\0';
2942 bs->backing_format[0] = '\0';
2943 bs->total_sectors = 0;
2944 bs->encrypted = false;
2945 bs->valid_key = false;
2946 bs->sg = false;
2947 QDECREF(bs->options);
2948 QDECREF(bs->explicit_options);
2949 bs->options = NULL;
2950 QDECREF(bs->full_open_options);
2951 bs->full_open_options = NULL;
2952 }
2953
2954 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2955 g_free(ban);
2956 }
2957 QLIST_INIT(&bs->aio_notifiers);
2958 bdrv_drained_end(bs);
2959 }
2960
2961 void bdrv_close_all(void)
2962 {
2963 block_job_cancel_sync_all();
2964 nbd_export_close_all();
2965
2966 /* Drop references from requests still in flight, such as canceled block
2967 * jobs whose AIO context has not been polled yet */
2968 bdrv_drain_all();
2969
2970 blk_remove_all_bs();
2971 blockdev_close_all_bdrv_states();
2972
2973 assert(QTAILQ_EMPTY(&all_bdrv_states));
2974 }
2975
2976 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
2977 {
2978 BdrvChild *to_c;
2979
2980 if (c->role->stay_at_node) {
2981 return false;
2982 }
2983
2984 if (c->role == &child_backing) {
2985 /* If @from is a backing file of @to, ignore the child to avoid
2986 * creating a loop. We only want to change the pointer of other
2987 * parents. */
2988 QLIST_FOREACH(to_c, &to->children, next) {
2989 if (to_c == c) {
2990 break;
2991 }
2992 }
2993 if (to_c) {
2994 return false;
2995 }
2996 }
2997
2998 return true;
2999 }
3000
3001 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3002 Error **errp)
3003 {
3004 BdrvChild *c, *next;
3005 GSList *list = NULL, *p;
3006 uint64_t old_perm, old_shared;
3007 uint64_t perm = 0, shared = BLK_PERM_ALL;
3008 int ret;
3009
3010 assert(!atomic_read(&from->in_flight));
3011 assert(!atomic_read(&to->in_flight));
3012
3013 /* Make sure that @from doesn't go away until we have successfully attached
3014 * all of its parents to @to. */
3015 bdrv_ref(from);
3016
3017 /* Put all parents into @list and calculate their cumulative permissions */
3018 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3019 if (!should_update_child(c, to)) {
3020 continue;
3021 }
3022 list = g_slist_prepend(list, c);
3023 perm |= c->perm;
3024 shared &= c->shared_perm;
3025 }
3026
3027 /* Check whether the required permissions can be granted on @to, ignoring
3028 * all BdrvChild in @list so that they can't block themselves. */
3029 ret = bdrv_check_update_perm(to, perm, shared, list, errp);
3030 if (ret < 0) {
3031 bdrv_abort_perm_update(to);
3032 goto out;
3033 }
3034
3035 /* Now actually perform the change. We performed the permission check for
3036 * all elements of @list at once, so set the permissions all at once at the
3037 * very end. */
3038 for (p = list; p != NULL; p = p->next) {
3039 c = p->data;
3040
3041 bdrv_ref(to);
3042 bdrv_replace_child_noperm(c, to);
3043 bdrv_unref(from);
3044 }
3045
3046 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3047 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3048
3049 out:
3050 g_slist_free(list);
3051 bdrv_unref(from);
3052 }
3053
3054 /*
3055 * Add new bs contents at the top of an image chain while the chain is
3056 * live, while keeping required fields on the top layer.
3057 *
3058 * This will modify the BlockDriverState fields, and swap contents
3059 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3060 *
3061 * bs_new must not be attached to a BlockBackend.
3062 *
3063 * This function does not create any image files.
3064 *
3065 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3066 * that's what the callers commonly need. bs_new will be referenced by the old
3067 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3068 * reference of its own, it must call bdrv_ref().
3069 */
3070 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3071 Error **errp)
3072 {
3073 Error *local_err = NULL;
3074
3075 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3076 if (local_err) {
3077 error_propagate(errp, local_err);
3078 goto out;
3079 }
3080
3081 bdrv_replace_node(bs_top, bs_new, &local_err);
3082 if (local_err) {
3083 error_propagate(errp, local_err);
3084 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3085 goto out;
3086 }
3087
3088 /* bs_new is now referenced by its new parents, we don't need the
3089 * additional reference any more. */
3090 out:
3091 bdrv_unref(bs_new);
3092 }
3093
3094 static void bdrv_delete(BlockDriverState *bs)
3095 {
3096 assert(!bs->job);
3097 assert(bdrv_op_blocker_is_empty(bs));
3098 assert(!bs->refcnt);
3099
3100 bdrv_close(bs);
3101
3102 /* remove from list, if necessary */
3103 if (bs->node_name[0] != '\0') {
3104 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3105 }
3106 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3107
3108 g_free(bs);
3109 }
3110
3111 /*
3112 * Run consistency checks on an image
3113 *
3114 * Returns 0 if the check could be completed (it doesn't mean that the image is
3115 * free of errors) or -errno when an internal error occurred. The results of the
3116 * check are stored in res.
3117 */
3118 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3119 {
3120 if (bs->drv == NULL) {
3121 return -ENOMEDIUM;
3122 }
3123 if (bs->drv->bdrv_check == NULL) {
3124 return -ENOTSUP;
3125 }
3126
3127 memset(res, 0, sizeof(*res));
3128 return bs->drv->bdrv_check(bs, res, fix);
3129 }
3130
3131 /*
3132 * Return values:
3133 * 0 - success
3134 * -EINVAL - backing format specified, but no file
3135 * -ENOSPC - can't update the backing file because no space is left in the
3136 * image file header
3137 * -ENOTSUP - format driver doesn't support changing the backing file
3138 */
3139 int bdrv_change_backing_file(BlockDriverState *bs,
3140 const char *backing_file, const char *backing_fmt)
3141 {
3142 BlockDriver *drv = bs->drv;
3143 int ret;
3144
3145 /* Backing file format doesn't make sense without a backing file */
3146 if (backing_fmt && !backing_file) {
3147 return -EINVAL;
3148 }
3149
3150 if (drv->bdrv_change_backing_file != NULL) {
3151 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3152 } else {
3153 ret = -ENOTSUP;
3154 }
3155
3156 if (ret == 0) {
3157 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3158 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3159 }
3160 return ret;
3161 }
3162
3163 /*
3164 * Finds the image layer in the chain that has 'bs' as its backing file.
3165 *
3166 * active is the current topmost image.
3167 *
3168 * Returns NULL if bs is not found in active's image chain,
3169 * or if active == bs.
3170 *
3171 * Returns the bottommost base image if bs == NULL.
3172 */
3173 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3174 BlockDriverState *bs)
3175 {
3176 while (active && bs != backing_bs(active)) {
3177 active = backing_bs(active);
3178 }
3179
3180 return active;
3181 }
3182
3183 /* Given a BDS, searches for the base layer. */
3184 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3185 {
3186 return bdrv_find_overlay(bs, NULL);
3187 }
3188
3189 /*
3190 * Drops images above 'base' up to and including 'top', and sets the image
3191 * above 'top' to have base as its backing file.
3192 *
3193 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3194 * information in 'bs' can be properly updated.
3195 *
3196 * E.g., this will convert the following chain:
3197 * bottom <- base <- intermediate <- top <- active
3198 *
3199 * to
3200 *
3201 * bottom <- base <- active
3202 *
3203 * It is allowed for bottom==base, in which case it converts:
3204 *
3205 * base <- intermediate <- top <- active
3206 *
3207 * to
3208 *
3209 * base <- active
3210 *
3211 * If backing_file_str is non-NULL, it will be used when modifying top's
3212 * overlay image metadata.
3213 *
3214 * Error conditions:
3215 * if active == top, that is considered an error
3216 *
3217 */
3218 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
3219 BlockDriverState *base, const char *backing_file_str)
3220 {
3221 BlockDriverState *new_top_bs = NULL;
3222 Error *local_err = NULL;
3223 int ret = -EIO;
3224
3225 if (!top->drv || !base->drv) {
3226 goto exit;
3227 }
3228
3229 new_top_bs = bdrv_find_overlay(active, top);
3230
3231 if (new_top_bs == NULL) {
3232 /* we could not find the image above 'top', this is an error */
3233 goto exit;
3234 }
3235
3236 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3237 * to do, no intermediate images */
3238 if (backing_bs(new_top_bs) == base) {
3239 ret = 0;
3240 goto exit;
3241 }
3242
3243 /* Make sure that base is in the backing chain of top */
3244 if (!bdrv_chain_contains(top, base)) {
3245 goto exit;
3246 }
3247
3248 /* success - we can delete the intermediate states, and link top->base */
3249 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3250 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
3251 base->drv ? base->drv->format_name : "");
3252 if (ret) {
3253 goto exit;
3254 }
3255
3256 bdrv_set_backing_hd(new_top_bs, base, &local_err);
3257 if (local_err) {
3258 ret = -EPERM;
3259 error_report_err(local_err);
3260 goto exit;
3261 }
3262
3263 ret = 0;
3264 exit:
3265 return ret;
3266 }
3267
3268 /**
3269 * Truncate file to 'offset' bytes (needed only for file protocols)
3270 */
3271 int bdrv_truncate(BdrvChild *child, int64_t offset)
3272 {
3273 BlockDriverState *bs = child->bs;
3274 BlockDriver *drv = bs->drv;
3275 int ret;
3276
3277 /* FIXME: Some format block drivers use this function instead of implicitly
3278 * growing their file by writing beyond its end.
3279 * See bdrv_aligned_pwritev() for an explanation why we currently
3280 * cannot assert this permission in that case. */
3281 // assert(child->perm & BLK_PERM_RESIZE);
3282
3283 if (!drv)
3284 return -ENOMEDIUM;
3285 if (!drv->bdrv_truncate)
3286 return -ENOTSUP;
3287 if (bs->read_only)
3288 return -EACCES;
3289
3290 ret = drv->bdrv_truncate(bs, offset);
3291 if (ret == 0) {
3292 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3293 bdrv_dirty_bitmap_truncate(bs);
3294 bdrv_parent_cb_resize(bs);
3295 ++bs->write_gen;
3296 }
3297 return ret;
3298 }
3299
3300 /**
3301 * Length of a allocated file in bytes. Sparse files are counted by actual
3302 * allocated space. Return < 0 if error or unknown.
3303 */
3304 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3305 {
3306 BlockDriver *drv = bs->drv;
3307 if (!drv) {
3308 return -ENOMEDIUM;
3309 }
3310 if (drv->bdrv_get_allocated_file_size) {
3311 return drv->bdrv_get_allocated_file_size(bs);
3312 }
3313 if (bs->file) {
3314 return bdrv_get_allocated_file_size(bs->file->bs);
3315 }
3316 return -ENOTSUP;
3317 }
3318
3319 /**
3320 * Return number of sectors on success, -errno on error.
3321 */
3322 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3323 {
3324 BlockDriver *drv = bs->drv;
3325
3326 if (!drv)
3327 return -ENOMEDIUM;
3328
3329 if (drv->has_variable_length) {
3330 int ret = refresh_total_sectors(bs, bs->total_sectors);
3331 if (ret < 0) {
3332 return ret;
3333 }
3334 }
3335 return bs->total_sectors;
3336 }
3337
3338 /**
3339 * Return length in bytes on success, -errno on error.
3340 * The length is always a multiple of BDRV_SECTOR_SIZE.
3341 */
3342 int64_t bdrv_getlength(BlockDriverState *bs)
3343 {
3344 int64_t ret = bdrv_nb_sectors(bs);
3345
3346 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3347 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3348 }
3349
3350 /* return 0 as number of sectors if no device present or error */
3351 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3352 {
3353 int64_t nb_sectors = bdrv_nb_sectors(bs);
3354
3355 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3356 }
3357
3358 bool bdrv_is_read_only(BlockDriverState *bs)
3359 {
3360 return bs->read_only;
3361 }
3362
3363 bool bdrv_is_sg(BlockDriverState *bs)
3364 {
3365 return bs->sg;
3366 }
3367
3368 bool bdrv_is_encrypted(BlockDriverState *bs)
3369 {
3370 if (bs->backing && bs->backing->bs->encrypted) {
3371 return true;
3372 }
3373 return bs->encrypted;
3374 }
3375
3376 bool bdrv_key_required(BlockDriverState *bs)
3377 {
3378 BdrvChild *backing = bs->backing;
3379
3380 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
3381 return true;
3382 }
3383 return (bs->encrypted && !bs->valid_key);
3384 }
3385
3386 int bdrv_set_key(BlockDriverState *bs, const char *key)
3387 {
3388 int ret;
3389 if (bs->backing && bs->backing->bs->encrypted) {
3390 ret = bdrv_set_key(bs->backing->bs, key);
3391 if (ret < 0)
3392 return ret;
3393 if (!bs->encrypted)
3394 return 0;
3395 }
3396 if (!bs->encrypted) {
3397 return -EINVAL;
3398 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3399 return -ENOMEDIUM;
3400 }
3401 ret = bs->drv->bdrv_set_key(bs, key);
3402 if (ret < 0) {
3403 bs->valid_key = false;
3404 } else if (!bs->valid_key) {
3405 /* call the change callback now, we skipped it on open */
3406 bs->valid_key = true;
3407 bdrv_parent_cb_change_media(bs, true);
3408 }
3409 return ret;
3410 }
3411
3412 /*
3413 * Provide an encryption key for @bs.
3414 * If @key is non-null:
3415 * If @bs is not encrypted, fail.
3416 * Else if the key is invalid, fail.
3417 * Else set @bs's key to @key, replacing the existing key, if any.
3418 * If @key is null:
3419 * If @bs is encrypted and still lacks a key, fail.
3420 * Else do nothing.
3421 * On failure, store an error object through @errp if non-null.
3422 */
3423 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
3424 {
3425 if (key) {
3426 if (!bdrv_is_encrypted(bs)) {
3427 error_setg(errp, "Node '%s' is not encrypted",
3428 bdrv_get_device_or_node_name(bs));
3429 } else if (bdrv_set_key(bs, key) < 0) {
3430 error_setg(errp, QERR_INVALID_PASSWORD);
3431 }
3432 } else {
3433 if (bdrv_key_required(bs)) {
3434 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
3435 "'%s' (%s) is encrypted",
3436 bdrv_get_device_or_node_name(bs),
3437 bdrv_get_encrypted_filename(bs));
3438 }
3439 }
3440 }
3441
3442 const char *bdrv_get_format_name(BlockDriverState *bs)
3443 {
3444 return bs->drv ? bs->drv->format_name : NULL;
3445 }
3446
3447 static int qsort_strcmp(const void *a, const void *b)
3448 {
3449 return strcmp(*(char *const *)a, *(char *const *)b);
3450 }
3451
3452 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3453 void *opaque)
3454 {
3455 BlockDriver *drv;
3456 int count = 0;
3457 int i;
3458 const char **formats = NULL;
3459
3460 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3461 if (drv->format_name) {
3462 bool found = false;
3463 int i = count;
3464 while (formats && i && !found) {
3465 found = !strcmp(formats[--i], drv->format_name);
3466 }
3467
3468 if (!found) {
3469 formats = g_renew(const char *, formats, count + 1);
3470 formats[count++] = drv->format_name;
3471 }
3472 }
3473 }
3474
3475 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3476 const char *format_name = block_driver_modules[i].format_name;
3477
3478 if (format_name) {
3479 bool found = false;
3480 int j = count;
3481
3482 while (formats && j && !found) {
3483 found = !strcmp(formats[--j], format_name);
3484 }
3485
3486 if (!found) {
3487 formats = g_renew(const char *, formats, count + 1);
3488 formats[count++] = format_name;
3489 }
3490 }
3491 }
3492
3493 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3494
3495 for (i = 0; i < count; i++) {
3496 it(opaque, formats[i]);
3497 }
3498
3499 g_free(formats);
3500 }
3501
3502 /* This function is to find a node in the bs graph */
3503 BlockDriverState *bdrv_find_node(const char *node_name)
3504 {
3505 BlockDriverState *bs;
3506
3507 assert(node_name);
3508
3509 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3510 if (!strcmp(node_name, bs->node_name)) {
3511 return bs;
3512 }
3513 }
3514 return NULL;
3515 }
3516
3517 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3518 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3519 {
3520 BlockDeviceInfoList *list, *entry;
3521 BlockDriverState *bs;
3522
3523 list = NULL;
3524 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3525 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3526 if (!info) {
3527 qapi_free_BlockDeviceInfoList(list);
3528 return NULL;
3529 }
3530 entry = g_malloc0(sizeof(*entry));
3531 entry->value = info;
3532 entry->next = list;
3533 list = entry;
3534 }
3535
3536 return list;
3537 }
3538
3539 BlockDriverState *bdrv_lookup_bs(const char *device,
3540 const char *node_name,
3541 Error **errp)
3542 {
3543 BlockBackend *blk;
3544 BlockDriverState *bs;
3545
3546 if (device) {
3547 blk = blk_by_name(device);
3548
3549 if (blk) {
3550 bs = blk_bs(blk);
3551 if (!bs) {
3552 error_setg(errp, "Device '%s' has no medium", device);
3553 }
3554
3555 return bs;
3556 }
3557 }
3558
3559 if (node_name) {
3560 bs = bdrv_find_node(node_name);
3561
3562 if (bs) {
3563 return bs;
3564 }
3565 }
3566
3567 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3568 device ? device : "",
3569 node_name ? node_name : "");
3570 return NULL;
3571 }
3572
3573 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3574 * return false. If either argument is NULL, return false. */
3575 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3576 {
3577 while (top && top != base) {
3578 top = backing_bs(top);
3579 }