block: Use BdrvChild callbacks for change_media/resize
[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 "trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "qemu/error-report.h"
29 #include "qemu/module.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qapi/qmp/qbool.h"
32 #include "qapi/qmp/qjson.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/sysemu.h"
35 #include "qemu/notify.h"
36 #include "qemu/coroutine.h"
37 #include "block/qapi.h"
38 #include "qmp-commands.h"
39 #include "qemu/timer.h"
40 #include "qapi-event.h"
41 #include "qemu/cutils.h"
42 #include "qemu/id.h"
43
44 #ifdef CONFIG_BSD
45 #include <sys/ioctl.h>
46 #include <sys/queue.h>
47 #ifndef __DragonFly__
48 #include <sys/disk.h>
49 #endif
50 #endif
51
52 #ifdef _WIN32
53 #include <windows.h>
54 #endif
55
56 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
57
58 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
59 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
60
61 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
63
64 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
65 QLIST_HEAD_INITIALIZER(bdrv_drivers);
66
67 static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
68 const char *reference, QDict *options, int flags,
69 BlockDriverState *parent,
70 const BdrvChildRole *child_role, Error **errp);
71
72 /* If non-zero, use only whitelisted block drivers */
73 static int use_bdrv_whitelist;
74
75 static void bdrv_close(BlockDriverState *bs);
76
77 #ifdef _WIN32
78 static int is_windows_drive_prefix(const char *filename)
79 {
80 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
81 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
82 filename[1] == ':');
83 }
84
85 int is_windows_drive(const char *filename)
86 {
87 if (is_windows_drive_prefix(filename) &&
88 filename[2] == '\0')
89 return 1;
90 if (strstart(filename, "\\\\.\\", NULL) ||
91 strstart(filename, "//./", NULL))
92 return 1;
93 return 0;
94 }
95 #endif
96
97 size_t bdrv_opt_mem_align(BlockDriverState *bs)
98 {
99 if (!bs || !bs->drv) {
100 /* page size or 4k (hdd sector size) should be on the safe side */
101 return MAX(4096, getpagesize());
102 }
103
104 return bs->bl.opt_mem_alignment;
105 }
106
107 size_t bdrv_min_mem_align(BlockDriverState *bs)
108 {
109 if (!bs || !bs->drv) {
110 /* page size or 4k (hdd sector size) should be on the safe side */
111 return MAX(4096, getpagesize());
112 }
113
114 return bs->bl.min_mem_alignment;
115 }
116
117 /* check if the path starts with "<protocol>:" */
118 int path_has_protocol(const char *path)
119 {
120 const char *p;
121
122 #ifdef _WIN32
123 if (is_windows_drive(path) ||
124 is_windows_drive_prefix(path)) {
125 return 0;
126 }
127 p = path + strcspn(path, ":/\\");
128 #else
129 p = path + strcspn(path, ":/");
130 #endif
131
132 return *p == ':';
133 }
134
135 int path_is_absolute(const char *path)
136 {
137 #ifdef _WIN32
138 /* specific case for names like: "\\.\d:" */
139 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
140 return 1;
141 }
142 return (*path == '/' || *path == '\\');
143 #else
144 return (*path == '/');
145 #endif
146 }
147
148 /* if filename is absolute, just copy it to dest. Otherwise, build a
149 path to it by considering it is relative to base_path. URL are
150 supported. */
151 void path_combine(char *dest, int dest_size,
152 const char *base_path,
153 const char *filename)
154 {
155 const char *p, *p1;
156 int len;
157
158 if (dest_size <= 0)
159 return;
160 if (path_is_absolute(filename)) {
161 pstrcpy(dest, dest_size, filename);
162 } else {
163 p = strchr(base_path, ':');
164 if (p)
165 p++;
166 else
167 p = base_path;
168 p1 = strrchr(base_path, '/');
169 #ifdef _WIN32
170 {
171 const char *p2;
172 p2 = strrchr(base_path, '\\');
173 if (!p1 || p2 > p1)
174 p1 = p2;
175 }
176 #endif
177 if (p1)
178 p1++;
179 else
180 p1 = base_path;
181 if (p1 > p)
182 p = p1;
183 len = p - base_path;
184 if (len > dest_size - 1)
185 len = dest_size - 1;
186 memcpy(dest, base_path, len);
187 dest[len] = '\0';
188 pstrcat(dest, dest_size, filename);
189 }
190 }
191
192 void bdrv_get_full_backing_filename_from_filename(const char *backed,
193 const char *backing,
194 char *dest, size_t sz,
195 Error **errp)
196 {
197 if (backing[0] == '\0' || path_has_protocol(backing) ||
198 path_is_absolute(backing))
199 {
200 pstrcpy(dest, sz, backing);
201 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
202 error_setg(errp, "Cannot use relative backing file names for '%s'",
203 backed);
204 } else {
205 path_combine(dest, sz, backed, backing);
206 }
207 }
208
209 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
210 Error **errp)
211 {
212 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
213
214 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
215 dest, sz, errp);
216 }
217
218 void bdrv_register(BlockDriver *bdrv)
219 {
220 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
221 }
222
223 BlockDriverState *bdrv_new_root(void)
224 {
225 return bdrv_new();
226 }
227
228 BlockDriverState *bdrv_new(void)
229 {
230 BlockDriverState *bs;
231 int i;
232
233 bs = g_new0(BlockDriverState, 1);
234 QLIST_INIT(&bs->dirty_bitmaps);
235 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
236 QLIST_INIT(&bs->op_blockers[i]);
237 }
238 notifier_with_return_list_init(&bs->before_write_notifiers);
239 bs->refcnt = 1;
240 bs->aio_context = qemu_get_aio_context();
241
242 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
243
244 return bs;
245 }
246
247 BlockDriver *bdrv_find_format(const char *format_name)
248 {
249 BlockDriver *drv1;
250 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
251 if (!strcmp(drv1->format_name, format_name)) {
252 return drv1;
253 }
254 }
255 return NULL;
256 }
257
258 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
259 {
260 static const char *whitelist_rw[] = {
261 CONFIG_BDRV_RW_WHITELIST
262 };
263 static const char *whitelist_ro[] = {
264 CONFIG_BDRV_RO_WHITELIST
265 };
266 const char **p;
267
268 if (!whitelist_rw[0] && !whitelist_ro[0]) {
269 return 1; /* no whitelist, anything goes */
270 }
271
272 for (p = whitelist_rw; *p; p++) {
273 if (!strcmp(drv->format_name, *p)) {
274 return 1;
275 }
276 }
277 if (read_only) {
278 for (p = whitelist_ro; *p; p++) {
279 if (!strcmp(drv->format_name, *p)) {
280 return 1;
281 }
282 }
283 }
284 return 0;
285 }
286
287 bool bdrv_uses_whitelist(void)
288 {
289 return use_bdrv_whitelist;
290 }
291
292 typedef struct CreateCo {
293 BlockDriver *drv;
294 char *filename;
295 QemuOpts *opts;
296 int ret;
297 Error *err;
298 } CreateCo;
299
300 static void coroutine_fn bdrv_create_co_entry(void *opaque)
301 {
302 Error *local_err = NULL;
303 int ret;
304
305 CreateCo *cco = opaque;
306 assert(cco->drv);
307
308 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
309 if (local_err) {
310 error_propagate(&cco->err, local_err);
311 }
312 cco->ret = ret;
313 }
314
315 int bdrv_create(BlockDriver *drv, const char* filename,
316 QemuOpts *opts, Error **errp)
317 {
318 int ret;
319
320 Coroutine *co;
321 CreateCo cco = {
322 .drv = drv,
323 .filename = g_strdup(filename),
324 .opts = opts,
325 .ret = NOT_DONE,
326 .err = NULL,
327 };
328
329 if (!drv->bdrv_create) {
330 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
331 ret = -ENOTSUP;
332 goto out;
333 }
334
335 if (qemu_in_coroutine()) {
336 /* Fast-path if already in coroutine context */
337 bdrv_create_co_entry(&cco);
338 } else {
339 co = qemu_coroutine_create(bdrv_create_co_entry);
340 qemu_coroutine_enter(co, &cco);
341 while (cco.ret == NOT_DONE) {
342 aio_poll(qemu_get_aio_context(), true);
343 }
344 }
345
346 ret = cco.ret;
347 if (ret < 0) {
348 if (cco.err) {
349 error_propagate(errp, cco.err);
350 } else {
351 error_setg_errno(errp, -ret, "Could not create image");
352 }
353 }
354
355 out:
356 g_free(cco.filename);
357 return ret;
358 }
359
360 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
361 {
362 BlockDriver *drv;
363 Error *local_err = NULL;
364 int ret;
365
366 drv = bdrv_find_protocol(filename, true, errp);
367 if (drv == NULL) {
368 return -ENOENT;
369 }
370
371 ret = bdrv_create(drv, filename, opts, &local_err);
372 if (local_err) {
373 error_propagate(errp, local_err);
374 }
375 return ret;
376 }
377
378 /**
379 * Try to get @bs's logical and physical block size.
380 * On success, store them in @bsz struct and return 0.
381 * On failure return -errno.
382 * @bs must not be empty.
383 */
384 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
385 {
386 BlockDriver *drv = bs->drv;
387
388 if (drv && drv->bdrv_probe_blocksizes) {
389 return drv->bdrv_probe_blocksizes(bs, bsz);
390 }
391
392 return -ENOTSUP;
393 }
394
395 /**
396 * Try to get @bs's geometry (cyls, heads, sectors).
397 * On success, store them in @geo struct and return 0.
398 * On failure return -errno.
399 * @bs must not be empty.
400 */
401 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
402 {
403 BlockDriver *drv = bs->drv;
404
405 if (drv && drv->bdrv_probe_geometry) {
406 return drv->bdrv_probe_geometry(bs, geo);
407 }
408
409 return -ENOTSUP;
410 }
411
412 /*
413 * Create a uniquely-named empty temporary file.
414 * Return 0 upon success, otherwise a negative errno value.
415 */
416 int get_tmp_filename(char *filename, int size)
417 {
418 #ifdef _WIN32
419 char temp_dir[MAX_PATH];
420 /* GetTempFileName requires that its output buffer (4th param)
421 have length MAX_PATH or greater. */
422 assert(size >= MAX_PATH);
423 return (GetTempPath(MAX_PATH, temp_dir)
424 && GetTempFileName(temp_dir, "qem", 0, filename)
425 ? 0 : -GetLastError());
426 #else
427 int fd;
428 const char *tmpdir;
429 tmpdir = getenv("TMPDIR");
430 if (!tmpdir) {
431 tmpdir = "/var/tmp";
432 }
433 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
434 return -EOVERFLOW;
435 }
436 fd = mkstemp(filename);
437 if (fd < 0) {
438 return -errno;
439 }
440 if (close(fd) != 0) {
441 unlink(filename);
442 return -errno;
443 }
444 return 0;
445 #endif
446 }
447
448 /*
449 * Detect host devices. By convention, /dev/cdrom[N] is always
450 * recognized as a host CDROM.
451 */
452 static BlockDriver *find_hdev_driver(const char *filename)
453 {
454 int score_max = 0, score;
455 BlockDriver *drv = NULL, *d;
456
457 QLIST_FOREACH(d, &bdrv_drivers, list) {
458 if (d->bdrv_probe_device) {
459 score = d->bdrv_probe_device(filename);
460 if (score > score_max) {
461 score_max = score;
462 drv = d;
463 }
464 }
465 }
466
467 return drv;
468 }
469
470 BlockDriver *bdrv_find_protocol(const char *filename,
471 bool allow_protocol_prefix,
472 Error **errp)
473 {
474 BlockDriver *drv1;
475 char protocol[128];
476 int len;
477 const char *p;
478
479 /* TODO Drivers without bdrv_file_open must be specified explicitly */
480
481 /*
482 * XXX(hch): we really should not let host device detection
483 * override an explicit protocol specification, but moving this
484 * later breaks access to device names with colons in them.
485 * Thanks to the brain-dead persistent naming schemes on udev-
486 * based Linux systems those actually are quite common.
487 */
488 drv1 = find_hdev_driver(filename);
489 if (drv1) {
490 return drv1;
491 }
492
493 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
494 return &bdrv_file;
495 }
496
497 p = strchr(filename, ':');
498 assert(p != NULL);
499 len = p - filename;
500 if (len > sizeof(protocol) - 1)
501 len = sizeof(protocol) - 1;
502 memcpy(protocol, filename, len);
503 protocol[len] = '\0';
504 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
505 if (drv1->protocol_name &&
506 !strcmp(drv1->protocol_name, protocol)) {
507 return drv1;
508 }
509 }
510
511 error_setg(errp, "Unknown protocol '%s'", protocol);
512 return NULL;
513 }
514
515 /*
516 * Guess image format by probing its contents.
517 * This is not a good idea when your image is raw (CVE-2008-2004), but
518 * we do it anyway for backward compatibility.
519 *
520 * @buf contains the image's first @buf_size bytes.
521 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
522 * but can be smaller if the image file is smaller)
523 * @filename is its filename.
524 *
525 * For all block drivers, call the bdrv_probe() method to get its
526 * probing score.
527 * Return the first block driver with the highest probing score.
528 */
529 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
530 const char *filename)
531 {
532 int score_max = 0, score;
533 BlockDriver *drv = NULL, *d;
534
535 QLIST_FOREACH(d, &bdrv_drivers, list) {
536 if (d->bdrv_probe) {
537 score = d->bdrv_probe(buf, buf_size, filename);
538 if (score > score_max) {
539 score_max = score;
540 drv = d;
541 }
542 }
543 }
544
545 return drv;
546 }
547
548 static int find_image_format(BlockDriverState *bs, const char *filename,
549 BlockDriver **pdrv, Error **errp)
550 {
551 BlockDriver *drv;
552 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
553 int ret = 0;
554
555 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
556 if (bdrv_is_sg(bs) || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
557 *pdrv = &bdrv_raw;
558 return ret;
559 }
560
561 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
562 if (ret < 0) {
563 error_setg_errno(errp, -ret, "Could not read image for determining its "
564 "format");
565 *pdrv = NULL;
566 return ret;
567 }
568
569 drv = bdrv_probe_all(buf, ret, filename);
570 if (!drv) {
571 error_setg(errp, "Could not determine image format: No compatible "
572 "driver found");
573 ret = -ENOENT;
574 }
575 *pdrv = drv;
576 return ret;
577 }
578
579 /**
580 * Set the current 'total_sectors' value
581 * Return 0 on success, -errno on error.
582 */
583 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
584 {
585 BlockDriver *drv = bs->drv;
586
587 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
588 if (bdrv_is_sg(bs))
589 return 0;
590
591 /* query actual device if possible, otherwise just trust the hint */
592 if (drv->bdrv_getlength) {
593 int64_t length = drv->bdrv_getlength(bs);
594 if (length < 0) {
595 return length;
596 }
597 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
598 }
599
600 bs->total_sectors = hint;
601 return 0;
602 }
603
604 /**
605 * Combines a QDict of new block driver @options with any missing options taken
606 * from @old_options, so that leaving out an option defaults to its old value.
607 */
608 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
609 QDict *old_options)
610 {
611 if (bs->drv && bs->drv->bdrv_join_options) {
612 bs->drv->bdrv_join_options(options, old_options);
613 } else {
614 qdict_join(options, old_options, false);
615 }
616 }
617
618 /**
619 * Set open flags for a given discard mode
620 *
621 * Return 0 on success, -1 if the discard mode was invalid.
622 */
623 int bdrv_parse_discard_flags(const char *mode, int *flags)
624 {
625 *flags &= ~BDRV_O_UNMAP;
626
627 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
628 /* do nothing */
629 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
630 *flags |= BDRV_O_UNMAP;
631 } else {
632 return -1;
633 }
634
635 return 0;
636 }
637
638 /**
639 * Set open flags for a given cache mode
640 *
641 * Return 0 on success, -1 if the cache mode was invalid.
642 */
643 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
644 {
645 *flags &= ~BDRV_O_CACHE_MASK;
646
647 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
648 *writethrough = false;
649 *flags |= BDRV_O_NOCACHE;
650 } else if (!strcmp(mode, "directsync")) {
651 *writethrough = true;
652 *flags |= BDRV_O_NOCACHE;
653 } else if (!strcmp(mode, "writeback")) {
654 *writethrough = false;
655 } else if (!strcmp(mode, "unsafe")) {
656 *writethrough = false;
657 *flags |= BDRV_O_NO_FLUSH;
658 } else if (!strcmp(mode, "writethrough")) {
659 *writethrough = true;
660 } else {
661 return -1;
662 }
663
664 return 0;
665 }
666
667 /*
668 * Returns the options and flags that a temporary snapshot should get, based on
669 * the originally requested flags (the originally requested image will have
670 * flags like a backing file)
671 */
672 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
673 int parent_flags, QDict *parent_options)
674 {
675 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
676
677 /* For temporary files, unconditional cache=unsafe is fine */
678 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
679 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
680 }
681
682 /*
683 * Returns the options and flags that bs->file should get if a protocol driver
684 * is expected, based on the given options and flags for the parent BDS
685 */
686 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
687 int parent_flags, QDict *parent_options)
688 {
689 int flags = parent_flags;
690
691 /* Enable protocol handling, disable format probing for bs->file */
692 flags |= BDRV_O_PROTOCOL;
693
694 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
695 * the parent. */
696 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
697 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
698
699 /* Our block drivers take care to send flushes and respect unmap policy,
700 * so we can default to enable both on lower layers regardless of the
701 * corresponding parent options. */
702 flags |= BDRV_O_UNMAP;
703
704 /* Clear flags that only apply to the top layer */
705 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
706 BDRV_O_NO_IO);
707
708 *child_flags = flags;
709 }
710
711 const BdrvChildRole child_file = {
712 .inherit_options = bdrv_inherited_options,
713 };
714
715 /*
716 * Returns the options and flags that bs->file should get if the use of formats
717 * (and not only protocols) is permitted for it, based on the given options and
718 * flags for the parent BDS
719 */
720 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
721 int parent_flags, QDict *parent_options)
722 {
723 child_file.inherit_options(child_flags, child_options,
724 parent_flags, parent_options);
725
726 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
727 }
728
729 const BdrvChildRole child_format = {
730 .inherit_options = bdrv_inherited_fmt_options,
731 };
732
733 /*
734 * Returns the options and flags that bs->backing should get, based on the
735 * given options and flags for the parent BDS
736 */
737 static void bdrv_backing_options(int *child_flags, QDict *child_options,
738 int parent_flags, QDict *parent_options)
739 {
740 int flags = parent_flags;
741
742 /* The cache mode is inherited unmodified for backing files; except WCE,
743 * which is only applied on the top level (BlockBackend) */
744 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
745 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
746
747 /* backing files always opened read-only */
748 flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
749
750 /* snapshot=on is handled on the top layer */
751 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
752
753 *child_flags = flags;
754 }
755
756 static const BdrvChildRole child_backing = {
757 .inherit_options = bdrv_backing_options,
758 };
759
760 static int bdrv_open_flags(BlockDriverState *bs, int flags)
761 {
762 int open_flags = flags;
763
764 /*
765 * Clear flags that are internal to the block layer before opening the
766 * image.
767 */
768 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
769
770 /*
771 * Snapshots should be writable.
772 */
773 if (flags & BDRV_O_TEMPORARY) {
774 open_flags |= BDRV_O_RDWR;
775 }
776
777 return open_flags;
778 }
779
780 static void update_flags_from_options(int *flags, QemuOpts *opts)
781 {
782 *flags &= ~BDRV_O_CACHE_MASK;
783
784 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
785 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
786 *flags |= BDRV_O_NO_FLUSH;
787 }
788
789 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
790 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
791 *flags |= BDRV_O_NOCACHE;
792 }
793 }
794
795 static void update_options_from_flags(QDict *options, int flags)
796 {
797 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
798 qdict_put(options, BDRV_OPT_CACHE_DIRECT,
799 qbool_from_bool(flags & BDRV_O_NOCACHE));
800 }
801 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
802 qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
803 qbool_from_bool(flags & BDRV_O_NO_FLUSH));
804 }
805 }
806
807 static void bdrv_assign_node_name(BlockDriverState *bs,
808 const char *node_name,
809 Error **errp)
810 {
811 char *gen_node_name = NULL;
812
813 if (!node_name) {
814 node_name = gen_node_name = id_generate(ID_BLOCK);
815 } else if (!id_wellformed(node_name)) {
816 /*
817 * Check for empty string or invalid characters, but not if it is
818 * generated (generated names use characters not available to the user)
819 */
820 error_setg(errp, "Invalid node name");
821 return;
822 }
823
824 /* takes care of avoiding namespaces collisions */
825 if (blk_by_name(node_name)) {
826 error_setg(errp, "node-name=%s is conflicting with a device id",
827 node_name);
828 goto out;
829 }
830
831 /* takes care of avoiding duplicates node names */
832 if (bdrv_find_node(node_name)) {
833 error_setg(errp, "Duplicate node name");
834 goto out;
835 }
836
837 /* copy node name into the bs and insert it into the graph list */
838 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
839 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
840 out:
841 g_free(gen_node_name);
842 }
843
844 static QemuOptsList bdrv_runtime_opts = {
845 .name = "bdrv_common",
846 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
847 .desc = {
848 {
849 .name = "node-name",
850 .type = QEMU_OPT_STRING,
851 .help = "Node name of the block device node",
852 },
853 {
854 .name = "driver",
855 .type = QEMU_OPT_STRING,
856 .help = "Block driver to use for the node",
857 },
858 {
859 .name = BDRV_OPT_CACHE_DIRECT,
860 .type = QEMU_OPT_BOOL,
861 .help = "Bypass software writeback cache on the host",
862 },
863 {
864 .name = BDRV_OPT_CACHE_NO_FLUSH,
865 .type = QEMU_OPT_BOOL,
866 .help = "Ignore flush requests",
867 },
868 { /* end of list */ }
869 },
870 };
871
872 /*
873 * Common part for opening disk images and files
874 *
875 * Removes all processed options from *options.
876 */
877 static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
878 QDict *options, Error **errp)
879 {
880 int ret, open_flags;
881 const char *filename;
882 const char *driver_name = NULL;
883 const char *node_name = NULL;
884 QemuOpts *opts;
885 BlockDriver *drv;
886 Error *local_err = NULL;
887
888 assert(bs->file == NULL);
889 assert(options != NULL && bs->options != options);
890
891 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
892 qemu_opts_absorb_qdict(opts, options, &local_err);
893 if (local_err) {
894 error_propagate(errp, local_err);
895 ret = -EINVAL;
896 goto fail_opts;
897 }
898
899 driver_name = qemu_opt_get(opts, "driver");
900 drv = bdrv_find_format(driver_name);
901 assert(drv != NULL);
902
903 if (file != NULL) {
904 filename = file->bs->filename;
905 } else {
906 filename = qdict_get_try_str(options, "filename");
907 }
908
909 if (drv->bdrv_needs_filename && !filename) {
910 error_setg(errp, "The '%s' block driver requires a file name",
911 drv->format_name);
912 ret = -EINVAL;
913 goto fail_opts;
914 }
915
916 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
917 drv->format_name);
918
919 node_name = qemu_opt_get(opts, "node-name");
920 bdrv_assign_node_name(bs, node_name, &local_err);
921 if (local_err) {
922 error_propagate(errp, local_err);
923 ret = -EINVAL;
924 goto fail_opts;
925 }
926
927 bs->request_alignment = 512;
928 bs->zero_beyond_eof = true;
929 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
930
931 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
932 error_setg(errp,
933 !bs->read_only && bdrv_is_whitelisted(drv, true)
934 ? "Driver '%s' can only be used for read-only devices"
935 : "Driver '%s' is not whitelisted",
936 drv->format_name);
937 ret = -ENOTSUP;
938 goto fail_opts;
939 }
940
941 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
942 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
943 if (!bs->read_only) {
944 bdrv_enable_copy_on_read(bs);
945 } else {
946 error_setg(errp, "Can't use copy-on-read on read-only device");
947 ret = -EINVAL;
948 goto fail_opts;
949 }
950 }
951
952 if (filename != NULL) {
953 pstrcpy(bs->filename, sizeof(bs->filename), filename);
954 } else {
955 bs->filename[0] = '\0';
956 }
957 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
958
959 bs->drv = drv;
960 bs->opaque = g_malloc0(drv->instance_size);
961
962 /* Apply cache mode options */
963 update_flags_from_options(&bs->open_flags, opts);
964
965 /* Open the image, either directly or using a protocol */
966 open_flags = bdrv_open_flags(bs, bs->open_flags);
967 if (drv->bdrv_file_open) {
968 assert(file == NULL);
969 assert(!drv->bdrv_needs_filename || filename != NULL);
970 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
971 } else {
972 if (file == NULL) {
973 error_setg(errp, "Can't use '%s' as a block driver for the "
974 "protocol level", drv->format_name);
975 ret = -EINVAL;
976 goto free_and_fail;
977 }
978 bs->file = file;
979 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
980 }
981
982 if (ret < 0) {
983 if (local_err) {
984 error_propagate(errp, local_err);
985 } else if (bs->filename[0]) {
986 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
987 } else {
988 error_setg_errno(errp, -ret, "Could not open image");
989 }
990 goto free_and_fail;
991 }
992
993 ret = refresh_total_sectors(bs, bs->total_sectors);
994 if (ret < 0) {
995 error_setg_errno(errp, -ret, "Could not refresh total sector count");
996 goto free_and_fail;
997 }
998
999 bdrv_refresh_limits(bs, &local_err);
1000 if (local_err) {
1001 error_propagate(errp, local_err);
1002 ret = -EINVAL;
1003 goto free_and_fail;
1004 }
1005
1006 assert(bdrv_opt_mem_align(bs) != 0);
1007 assert(bdrv_min_mem_align(bs) != 0);
1008 assert((bs->request_alignment != 0) || bdrv_is_sg(bs));
1009
1010 qemu_opts_del(opts);
1011 return 0;
1012
1013 free_and_fail:
1014 bs->file = NULL;
1015 g_free(bs->opaque);
1016 bs->opaque = NULL;
1017 bs->drv = NULL;
1018 fail_opts:
1019 qemu_opts_del(opts);
1020 return ret;
1021 }
1022
1023 static QDict *parse_json_filename(const char *filename, Error **errp)
1024 {
1025 QObject *options_obj;
1026 QDict *options;
1027 int ret;
1028
1029 ret = strstart(filename, "json:", &filename);
1030 assert(ret);
1031
1032 options_obj = qobject_from_json(filename);
1033 if (!options_obj) {
1034 error_setg(errp, "Could not parse the JSON options");
1035 return NULL;
1036 }
1037
1038 if (qobject_type(options_obj) != QTYPE_QDICT) {
1039 qobject_decref(options_obj);
1040 error_setg(errp, "Invalid JSON object given");
1041 return NULL;
1042 }
1043
1044 options = qobject_to_qdict(options_obj);
1045 qdict_flatten(options);
1046
1047 return options;
1048 }
1049
1050 static void parse_json_protocol(QDict *options, const char **pfilename,
1051 Error **errp)
1052 {
1053 QDict *json_options;
1054 Error *local_err = NULL;
1055
1056 /* Parse json: pseudo-protocol */
1057 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1058 return;
1059 }
1060
1061 json_options = parse_json_filename(*pfilename, &local_err);
1062 if (local_err) {
1063 error_propagate(errp, local_err);
1064 return;
1065 }
1066
1067 /* Options given in the filename have lower priority than options
1068 * specified directly */
1069 qdict_join(options, json_options, false);
1070 QDECREF(json_options);
1071 *pfilename = NULL;
1072 }
1073
1074 /*
1075 * Fills in default options for opening images and converts the legacy
1076 * filename/flags pair to option QDict entries.
1077 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1078 * block driver has been specified explicitly.
1079 */
1080 static int bdrv_fill_options(QDict **options, const char *filename,
1081 int *flags, Error **errp)
1082 {
1083 const char *drvname;
1084 bool protocol = *flags & BDRV_O_PROTOCOL;
1085 bool parse_filename = false;
1086 BlockDriver *drv = NULL;
1087 Error *local_err = NULL;
1088
1089 drvname = qdict_get_try_str(*options, "driver");
1090 if (drvname) {
1091 drv = bdrv_find_format(drvname);
1092 if (!drv) {
1093 error_setg(errp, "Unknown driver '%s'", drvname);
1094 return -ENOENT;
1095 }
1096 /* If the user has explicitly specified the driver, this choice should
1097 * override the BDRV_O_PROTOCOL flag */
1098 protocol = drv->bdrv_file_open;
1099 }
1100
1101 if (protocol) {
1102 *flags |= BDRV_O_PROTOCOL;
1103 } else {
1104 *flags &= ~BDRV_O_PROTOCOL;
1105 }
1106
1107 /* Translate cache options from flags into options */
1108 update_options_from_flags(*options, *flags);
1109
1110 /* Fetch the file name from the options QDict if necessary */
1111 if (protocol && filename) {
1112 if (!qdict_haskey(*options, "filename")) {
1113 qdict_put(*options, "filename", qstring_from_str(filename));
1114 parse_filename = true;
1115 } else {
1116 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1117 "the same time");
1118 return -EINVAL;
1119 }
1120 }
1121
1122 /* Find the right block driver */
1123 filename = qdict_get_try_str(*options, "filename");
1124
1125 if (!drvname && protocol) {
1126 if (filename) {
1127 drv = bdrv_find_protocol(filename, parse_filename, errp);
1128 if (!drv) {
1129 return -EINVAL;
1130 }
1131
1132 drvname = drv->format_name;
1133 qdict_put(*options, "driver", qstring_from_str(drvname));
1134 } else {
1135 error_setg(errp, "Must specify either driver or file");
1136 return -EINVAL;
1137 }
1138 }
1139
1140 assert(drv || !protocol);
1141
1142 /* Driver-specific filename parsing */
1143 if (drv && drv->bdrv_parse_filename && parse_filename) {
1144 drv->bdrv_parse_filename(filename, *options, &local_err);
1145 if (local_err) {
1146 error_propagate(errp, local_err);
1147 return -EINVAL;
1148 }
1149
1150 if (!drv->bdrv_needs_filename) {
1151 qdict_del(*options, "filename");
1152 }
1153 }
1154
1155 return 0;
1156 }
1157
1158 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1159 const char *child_name,
1160 const BdrvChildRole *child_role)
1161 {
1162 BdrvChild *child = g_new(BdrvChild, 1);
1163 *child = (BdrvChild) {
1164 .bs = child_bs,
1165 .name = g_strdup(child_name),
1166 .role = child_role,
1167 };
1168
1169 QLIST_INSERT_HEAD(&child_bs->parents, child, next_parent);
1170
1171 return child;
1172 }
1173
1174 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1175 BlockDriverState *child_bs,
1176 const char *child_name,
1177 const BdrvChildRole *child_role)
1178 {
1179 BdrvChild *child = bdrv_root_attach_child(child_bs, child_name, child_role);
1180 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1181 return child;
1182 }
1183
1184 static void bdrv_detach_child(BdrvChild *child)
1185 {
1186 if (child->next.le_prev) {
1187 QLIST_REMOVE(child, next);
1188 child->next.le_prev = NULL;
1189 }
1190 QLIST_REMOVE(child, next_parent);
1191 g_free(child->name);
1192 g_free(child);
1193 }
1194
1195 void bdrv_root_unref_child(BdrvChild *child)
1196 {
1197 BlockDriverState *child_bs;
1198
1199 child_bs = child->bs;
1200 bdrv_detach_child(child);
1201 bdrv_unref(child_bs);
1202 }
1203
1204 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1205 {
1206 if (child == NULL) {
1207 return;
1208 }
1209
1210 if (child->bs->inherits_from == parent) {
1211 child->bs->inherits_from = NULL;
1212 }
1213
1214 bdrv_root_unref_child(child);
1215 }
1216
1217
1218 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1219 {
1220 BdrvChild *c;
1221 QLIST_FOREACH(c, &bs->parents, next_parent) {
1222 if (c->role->change_media) {
1223 c->role->change_media(c, load);
1224 }
1225 }
1226 }
1227
1228 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1229 {
1230 BdrvChild *c;
1231 QLIST_FOREACH(c, &bs->parents, next_parent) {
1232 if (c->role->resize) {
1233 c->role->resize(c);
1234 }
1235 }
1236 }
1237
1238 /*
1239 * Sets the backing file link of a BDS. A new reference is created; callers
1240 * which don't need their own reference any more must call bdrv_unref().
1241 */
1242 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1243 {
1244 if (backing_hd) {
1245 bdrv_ref(backing_hd);
1246 }
1247
1248 if (bs->backing) {
1249 assert(bs->backing_blocker);
1250 bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker);
1251 bdrv_unref_child(bs, bs->backing);
1252 } else if (backing_hd) {
1253 error_setg(&bs->backing_blocker,
1254 "node is used as backing hd of '%s'",
1255 bdrv_get_device_or_node_name(bs));
1256 }
1257
1258 if (!backing_hd) {
1259 error_free(bs->backing_blocker);
1260 bs->backing_blocker = NULL;
1261 bs->backing = NULL;
1262 goto out;
1263 }
1264 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing);
1265 bs->open_flags &= ~BDRV_O_NO_BACKING;
1266 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1267 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1268 backing_hd->drv ? backing_hd->drv->format_name : "");
1269
1270 bdrv_op_block_all(backing_hd, bs->backing_blocker);
1271 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1272 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1273 bs->backing_blocker);
1274 out:
1275 bdrv_refresh_limits(bs, NULL);
1276 }
1277
1278 /*
1279 * Opens the backing file for a BlockDriverState if not yet open
1280 *
1281 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1282 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1283 * itself, all options starting with "${bdref_key}." are considered part of the
1284 * BlockdevRef.
1285 *
1286 * TODO Can this be unified with bdrv_open_image()?
1287 */
1288 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1289 const char *bdref_key, Error **errp)
1290 {
1291 char *backing_filename = g_malloc0(PATH_MAX);
1292 char *bdref_key_dot;
1293 const char *reference = NULL;
1294 int ret = 0;
1295 BlockDriverState *backing_hd;
1296 QDict *options;
1297 QDict *tmp_parent_options = NULL;
1298 Error *local_err = NULL;
1299
1300 if (bs->backing != NULL) {
1301 goto free_exit;
1302 }
1303
1304 /* NULL means an empty set of options */
1305 if (parent_options == NULL) {
1306 tmp_parent_options = qdict_new();
1307 parent_options = tmp_parent_options;
1308 }
1309
1310 bs->open_flags &= ~BDRV_O_NO_BACKING;
1311
1312 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1313 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
1314 g_free(bdref_key_dot);
1315
1316 reference = qdict_get_try_str(parent_options, bdref_key);
1317 if (reference || qdict_haskey(options, "file.filename")) {
1318 backing_filename[0] = '\0';
1319 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1320 QDECREF(options);
1321 goto free_exit;
1322 } else {
1323 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
1324 &local_err);
1325 if (local_err) {
1326 ret = -EINVAL;
1327 error_propagate(errp, local_err);
1328 QDECREF(options);
1329 goto free_exit;
1330 }
1331 }
1332
1333 if (!bs->drv || !bs->drv->supports_backing) {
1334 ret = -EINVAL;
1335 error_setg(errp, "Driver doesn't support backing files");
1336 QDECREF(options);
1337 goto free_exit;
1338 }
1339
1340 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
1341 qdict_put(options, "driver", qstring_from_str(bs->backing_format));
1342 }
1343
1344 backing_hd = NULL;
1345 ret = bdrv_open_inherit(&backing_hd,
1346 *backing_filename ? backing_filename : NULL,
1347 reference, options, 0, bs, &child_backing,
1348 errp);
1349 if (ret < 0) {
1350 bs->open_flags |= BDRV_O_NO_BACKING;
1351 error_prepend(errp, "Could not open backing file: ");
1352 goto free_exit;
1353 }
1354
1355 /* Hook up the backing file link; drop our reference, bs owns the
1356 * backing_hd reference now */
1357 bdrv_set_backing_hd(bs, backing_hd);
1358 bdrv_unref(backing_hd);
1359
1360 qdict_del(parent_options, bdref_key);
1361
1362 free_exit:
1363 g_free(backing_filename);
1364 QDECREF(tmp_parent_options);
1365 return ret;
1366 }
1367
1368 /*
1369 * Opens a disk image whose options are given as BlockdevRef in another block
1370 * device's options.
1371 *
1372 * If allow_none is true, no image will be opened if filename is false and no
1373 * BlockdevRef is given. NULL will be returned, but errp remains unset.
1374 *
1375 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1376 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1377 * itself, all options starting with "${bdref_key}." are considered part of the
1378 * BlockdevRef.
1379 *
1380 * The BlockdevRef will be removed from the options QDict.
1381 */
1382 BdrvChild *bdrv_open_child(const char *filename,
1383 QDict *options, const char *bdref_key,
1384 BlockDriverState* parent,
1385 const BdrvChildRole *child_role,
1386 bool allow_none, Error **errp)
1387 {
1388 BdrvChild *c = NULL;
1389 BlockDriverState *bs;
1390 QDict *image_options;
1391 int ret;
1392 char *bdref_key_dot;
1393 const char *reference;
1394
1395 assert(child_role != NULL);
1396
1397 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1398 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1399 g_free(bdref_key_dot);
1400
1401 reference = qdict_get_try_str(options, bdref_key);
1402 if (!filename && !reference && !qdict_size(image_options)) {
1403 if (!allow_none) {
1404 error_setg(errp, "A block device must be specified for \"%s\"",
1405 bdref_key);
1406 }
1407 QDECREF(image_options);
1408 goto done;
1409 }
1410
1411 bs = NULL;
1412 ret = bdrv_open_inherit(&bs, filename, reference, image_options, 0,
1413 parent, child_role, errp);
1414 if (ret < 0) {
1415 goto done;
1416 }
1417
1418 c = bdrv_attach_child(parent, bs, bdref_key, child_role);
1419
1420 done:
1421 qdict_del(options, bdref_key);
1422 return c;
1423 }
1424
1425 static int bdrv_append_temp_snapshot(BlockDriverState *bs, int flags,
1426 QDict *snapshot_options, Error **errp)
1427 {
1428 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1429 char *tmp_filename = g_malloc0(PATH_MAX + 1);
1430 int64_t total_size;
1431 QemuOpts *opts = NULL;
1432 BlockDriverState *bs_snapshot;
1433 Error *local_err = NULL;
1434 int ret;
1435
1436 /* if snapshot, we create a temporary backing file and open it
1437 instead of opening 'filename' directly */
1438
1439 /* Get the required size from the image */
1440 total_size = bdrv_getlength(bs);
1441 if (total_size < 0) {
1442 ret = total_size;
1443 error_setg_errno(errp, -total_size, "Could not get image size");
1444 goto out;
1445 }
1446
1447 /* Create the temporary image */
1448 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1449 if (ret < 0) {
1450 error_setg_errno(errp, -ret, "Could not get temporary filename");
1451 goto out;
1452 }
1453
1454 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
1455 &error_abort);
1456 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
1457 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
1458 qemu_opts_del(opts);
1459 if (ret < 0) {
1460 error_prepend(errp, "Could not create temporary overlay '%s': ",
1461 tmp_filename);
1462 goto out;
1463 }
1464
1465 /* Prepare options QDict for the temporary file */
1466 qdict_put(snapshot_options, "file.driver",
1467 qstring_from_str("file"));
1468 qdict_put(snapshot_options, "file.filename",
1469 qstring_from_str(tmp_filename));
1470 qdict_put(snapshot_options, "driver",
1471 qstring_from_str("qcow2"));
1472
1473 bs_snapshot = bdrv_new();
1474
1475 ret = bdrv_open(&bs_snapshot, NULL, NULL, snapshot_options,
1476 flags, &local_err);
1477 snapshot_options = NULL;
1478 if (ret < 0) {
1479 error_propagate(errp, local_err);
1480 goto out;
1481 }
1482
1483 bdrv_append(bs_snapshot, bs);
1484
1485 out:
1486 QDECREF(snapshot_options);
1487 g_free(tmp_filename);
1488 return ret;
1489 }
1490
1491 /*
1492 * Opens a disk image (raw, qcow2, vmdk, ...)
1493 *
1494 * options is a QDict of options to pass to the block drivers, or NULL for an
1495 * empty set of options. The reference to the QDict belongs to the block layer
1496 * after the call (even on failure), so if the caller intends to reuse the
1497 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1498 *
1499 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1500 * If it is not NULL, the referenced BDS will be reused.
1501 *
1502 * The reference parameter may be used to specify an existing block device which
1503 * should be opened. If specified, neither options nor a filename may be given,
1504 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1505 */
1506 static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
1507 const char *reference, QDict *options, int flags,
1508 BlockDriverState *parent,
1509 const BdrvChildRole *child_role, Error **errp)
1510 {
1511 int ret;
1512 BdrvChild *file = NULL;
1513 BlockDriverState *bs;
1514 BlockDriver *drv = NULL;
1515 const char *drvname;
1516 const char *backing;
1517 Error *local_err = NULL;
1518 QDict *snapshot_options = NULL;
1519 int snapshot_flags = 0;
1520
1521 assert(pbs);
1522 assert(!child_role || !flags);
1523 assert(!child_role == !parent);
1524
1525 if (reference) {
1526 bool options_non_empty = options ? qdict_size(options) : false;
1527 QDECREF(options);
1528
1529 if (*pbs) {
1530 error_setg(errp, "Cannot reuse an existing BDS when referencing "
1531 "another block device");
1532 return -EINVAL;
1533 }
1534
1535 if (filename || options_non_empty) {
1536 error_setg(errp, "Cannot reference an existing block device with "
1537 "additional options or a new filename");
1538 return -EINVAL;
1539 }
1540
1541 bs = bdrv_lookup_bs(reference, reference, errp);
1542 if (!bs) {
1543 return -ENODEV;
1544 }
1545
1546 bdrv_ref(bs);
1547 *pbs = bs;
1548 return 0;
1549 }
1550
1551 if (*pbs) {
1552 bs = *pbs;
1553 } else {
1554 bs = bdrv_new();
1555 }
1556
1557 /* NULL means an empty set of options */
1558 if (options == NULL) {
1559 options = qdict_new();
1560 }
1561
1562 /* json: syntax counts as explicit options, as if in the QDict */
1563 parse_json_protocol(options, &filename, &local_err);
1564 if (local_err) {
1565 ret = -EINVAL;
1566 goto fail;
1567 }
1568
1569 bs->explicit_options = qdict_clone_shallow(options);
1570
1571 if (child_role) {
1572 bs->inherits_from = parent;
1573 child_role->inherit_options(&flags, options,
1574 parent->open_flags, parent->options);
1575 }
1576
1577 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
1578 if (local_err) {
1579 goto fail;
1580 }
1581
1582 bs->open_flags = flags;
1583 bs->options = options;
1584 options = qdict_clone_shallow(options);
1585
1586 /* Find the right image format driver */
1587 drvname = qdict_get_try_str(options, "driver");
1588 if (drvname) {
1589 drv = bdrv_find_format(drvname);
1590 if (!drv) {
1591 error_setg(errp, "Unknown driver: '%s'", drvname);
1592 ret = -EINVAL;
1593 goto fail;
1594 }
1595 }
1596
1597 assert(drvname || !(flags & BDRV_O_PROTOCOL));
1598
1599 backing = qdict_get_try_str(options, "backing");
1600 if (backing && *backing == '\0') {
1601 flags |= BDRV_O_NO_BACKING;
1602 qdict_del(options, "backing");
1603 }
1604
1605 /* Open image file without format layer */
1606 if ((flags & BDRV_O_PROTOCOL) == 0) {
1607 if (flags & BDRV_O_RDWR) {
1608 flags |= BDRV_O_ALLOW_RDWR;
1609 }
1610 if (flags & BDRV_O_SNAPSHOT) {
1611 snapshot_options = qdict_new();
1612 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
1613 flags, options);
1614 bdrv_backing_options(&flags, options, flags, options);
1615 }
1616
1617 bs->open_flags = flags;
1618
1619 file = bdrv_open_child(filename, options, "file", bs,
1620 &child_file, true, &local_err);
1621 if (local_err) {
1622 ret = -EINVAL;
1623 goto fail;
1624 }
1625 }
1626
1627 /* Image format probing */
1628 bs->probed = !drv;
1629 if (!drv && file) {
1630 ret = find_image_format(file->bs, filename, &drv, &local_err);
1631 if (ret < 0) {
1632 goto fail;
1633 }
1634 /*
1635 * This option update would logically belong in bdrv_fill_options(),
1636 * but we first need to open bs->file for the probing to work, while
1637 * opening bs->file already requires the (mostly) final set of options
1638 * so that cache mode etc. can be inherited.
1639 *
1640 * Adding the driver later is somewhat ugly, but it's not an option
1641 * that would ever be inherited, so it's correct. We just need to make
1642 * sure to update both bs->options (which has the full effective
1643 * options for bs) and options (which has file.* already removed).
1644 */
1645 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
1646 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1647 } else if (!drv) {
1648 error_setg(errp, "Must specify either driver or file");
1649 ret = -EINVAL;
1650 goto fail;
1651 }
1652
1653 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
1654 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
1655 /* file must be NULL if a protocol BDS is about to be created
1656 * (the inverse results in an error message from bdrv_open_common()) */
1657 assert(!(flags & BDRV_O_PROTOCOL) || !file);
1658
1659 /* Open the image */
1660 ret = bdrv_open_common(bs, file, options, &local_err);
1661 if (ret < 0) {
1662 goto fail;
1663 }
1664
1665 if (file && (bs->file != file)) {
1666 bdrv_unref_child(bs, file);
1667 file = NULL;
1668 }
1669
1670 /* If there is a backing file, use it */
1671 if ((flags & BDRV_O_NO_BACKING) == 0) {
1672 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
1673 if (ret < 0) {
1674 goto close_and_fail;
1675 }
1676 }
1677
1678 bdrv_refresh_filename(bs);
1679
1680 /* Check if any unknown options were used */
1681 if (options && (qdict_size(options) != 0)) {
1682 const QDictEntry *entry = qdict_first(options);
1683 if (flags & BDRV_O_PROTOCOL) {
1684 error_setg(errp, "Block protocol '%s' doesn't support the option "
1685 "'%s'", drv->format_name, entry->key);
1686 } else {
1687 error_setg(errp,
1688 "Block format '%s' does not support the option '%s'",
1689 drv->format_name, entry->key);
1690 }
1691
1692 ret = -EINVAL;
1693 goto close_and_fail;
1694 }
1695
1696 if (!bdrv_key_required(bs)) {
1697 bdrv_parent_cb_change_media(bs, true);
1698 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1699 && !runstate_check(RUN_STATE_INMIGRATE)
1700 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1701 error_setg(errp,
1702 "Guest must be stopped for opening of encrypted image");
1703 ret = -EBUSY;
1704 goto close_and_fail;
1705 }
1706
1707 QDECREF(options);
1708 *pbs = bs;
1709
1710 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1711 * temporary snapshot afterwards. */
1712 if (snapshot_flags) {
1713 ret = bdrv_append_temp_snapshot(bs, snapshot_flags, snapshot_options,
1714 &local_err);
1715 snapshot_options = NULL;
1716 if (local_err) {
1717 goto close_and_fail;
1718 }
1719 }
1720
1721 return 0;
1722
1723 fail:
1724 if (file != NULL) {
1725 bdrv_unref_child(bs, file);
1726 }
1727 QDECREF(snapshot_options);
1728 QDECREF(bs->explicit_options);
1729 QDECREF(bs->options);
1730 QDECREF(options);
1731 bs->options = NULL;
1732 if (!*pbs) {
1733 /* If *pbs is NULL, a new BDS has been created in this function and
1734 needs to be freed now. Otherwise, it does not need to be closed,
1735 since it has not really been opened yet. */
1736 bdrv_unref(bs);
1737 }
1738 if (local_err) {
1739 error_propagate(errp, local_err);
1740 }
1741 return ret;
1742
1743 close_and_fail:
1744 /* See fail path, but now the BDS has to be always closed */
1745 if (*pbs) {
1746 bdrv_close(bs);
1747 } else {
1748 bdrv_unref(bs);
1749 }
1750 QDECREF(snapshot_options);
1751 QDECREF(options);
1752 if (local_err) {
1753 error_propagate(errp, local_err);
1754 }
1755 return ret;
1756 }
1757
1758 int bdrv_open(BlockDriverState **pbs, const char *filename,
1759 const char *reference, QDict *options, int flags, Error **errp)
1760 {
1761 return bdrv_open_inherit(pbs, filename, reference, options, flags, NULL,
1762 NULL, errp);
1763 }
1764
1765 typedef struct BlockReopenQueueEntry {
1766 bool prepared;
1767 BDRVReopenState state;
1768 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1769 } BlockReopenQueueEntry;
1770
1771 /*
1772 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1773 * reopen of multiple devices.
1774 *
1775 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1776 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1777 * be created and initialized. This newly created BlockReopenQueue should be
1778 * passed back in for subsequent calls that are intended to be of the same
1779 * atomic 'set'.
1780 *
1781 * bs is the BlockDriverState to add to the reopen queue.
1782 *
1783 * options contains the changed options for the associated bs
1784 * (the BlockReopenQueue takes ownership)
1785 *
1786 * flags contains the open flags for the associated bs
1787 *
1788 * returns a pointer to bs_queue, which is either the newly allocated
1789 * bs_queue, or the existing bs_queue being used.
1790 *
1791 */
1792 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
1793 BlockDriverState *bs,
1794 QDict *options,
1795 int flags,
1796 const BdrvChildRole *role,
1797 QDict *parent_options,
1798 int parent_flags)
1799 {
1800 assert(bs != NULL);
1801
1802 BlockReopenQueueEntry *bs_entry;
1803 BdrvChild *child;
1804 QDict *old_options, *explicit_options;
1805
1806 if (bs_queue == NULL) {
1807 bs_queue = g_new0(BlockReopenQueue, 1);
1808 QSIMPLEQ_INIT(bs_queue);
1809 }
1810
1811 if (!options) {
1812 options = qdict_new();
1813 }
1814
1815 /*
1816 * Precedence of options:
1817 * 1. Explicitly passed in options (highest)
1818 * 2. Set in flags (only for top level)
1819 * 3. Retained from explicitly set options of bs
1820 * 4. Inherited from parent node
1821 * 5. Retained from effective options of bs
1822 */
1823
1824 if (!parent_options) {
1825 /*
1826 * Any setting represented by flags is always updated. If the
1827 * corresponding QDict option is set, it takes precedence. Otherwise
1828 * the flag is translated into a QDict option. The old setting of bs is
1829 * not considered.
1830 */
1831 update_options_from_flags(options, flags);
1832 }
1833
1834 /* Old explicitly set values (don't overwrite by inherited value) */
1835 old_options = qdict_clone_shallow(bs->explicit_options);
1836 bdrv_join_options(bs, options, old_options);
1837 QDECREF(old_options);
1838
1839 explicit_options = qdict_clone_shallow(options);
1840
1841 /* Inherit from parent node */
1842 if (parent_options) {
1843 assert(!flags);
1844 role->inherit_options(&flags, options, parent_flags, parent_options);
1845 }
1846
1847 /* Old values are used for options that aren't set yet */
1848 old_options = qdict_clone_shallow(bs->options);
1849 bdrv_join_options(bs, options, old_options);
1850 QDECREF(old_options);
1851
1852 /* bdrv_open() masks this flag out */
1853 flags &= ~BDRV_O_PROTOCOL;
1854
1855 QLIST_FOREACH(child, &bs->children, next) {
1856 QDict *new_child_options;
1857 char *child_key_dot;
1858
1859 /* reopen can only change the options of block devices that were
1860 * implicitly created and inherited options. For other (referenced)
1861 * block devices, a syntax like "backing.foo" results in an error. */
1862 if (child->bs->inherits_from != bs) {
1863 continue;
1864 }
1865
1866 child_key_dot = g_strdup_printf("%s.", child->name);
1867 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
1868 g_free(child_key_dot);
1869
1870 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
1871 child->role, options, flags);
1872 }
1873
1874 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1875 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1876
1877 bs_entry->state.bs = bs;
1878 bs_entry->state.options = options;
1879 bs_entry->state.explicit_options = explicit_options;
1880 bs_entry->state.flags = flags;
1881
1882 return bs_queue;
1883 }
1884
1885 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1886 BlockDriverState *bs,
1887 QDict *options, int flags)
1888 {
1889 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
1890 NULL, NULL, 0);
1891 }
1892
1893 /*
1894 * Reopen multiple BlockDriverStates atomically & transactionally.
1895 *
1896 * The queue passed in (bs_queue) must have been built up previous
1897 * via bdrv_reopen_queue().
1898 *
1899 * Reopens all BDS specified in the queue, with the appropriate
1900 * flags. All devices are prepared for reopen, and failure of any
1901 * device will cause all device changes to be abandonded, and intermediate
1902 * data cleaned up.
1903 *
1904 * If all devices prepare successfully, then the changes are committed
1905 * to all devices.
1906 *
1907 */
1908 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1909 {
1910 int ret = -1;
1911 BlockReopenQueueEntry *bs_entry, *next;
1912 Error *local_err = NULL;
1913
1914 assert(bs_queue != NULL);
1915
1916 bdrv_drain_all();
1917
1918 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1919 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1920 error_propagate(errp, local_err);
1921 goto cleanup;
1922 }
1923 bs_entry->prepared = true;
1924 }
1925
1926 /* If we reach this point, we have success and just need to apply the
1927 * changes
1928 */
1929 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1930 bdrv_reopen_commit(&bs_entry->state);
1931 }
1932
1933 ret = 0;
1934
1935 cleanup:
1936 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1937 if (ret && bs_entry->prepared) {
1938 bdrv_reopen_abort(&bs_entry->state);
1939 } else if (ret) {
1940 QDECREF(bs_entry->state.explicit_options);
1941 }
1942 QDECREF(bs_entry->state.options);
1943 g_free(bs_entry);
1944 }
1945 g_free(bs_queue);
1946 return ret;
1947 }
1948
1949
1950 /* Reopen a single BlockDriverState with the specified flags. */
1951 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1952 {
1953 int ret = -1;
1954 Error *local_err = NULL;
1955 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
1956
1957 ret = bdrv_reopen_multiple(queue, &local_err);
1958 if (local_err != NULL) {
1959 error_propagate(errp, local_err);
1960 }
1961 return ret;
1962 }
1963
1964
1965 /*
1966 * Prepares a BlockDriverState for reopen. All changes are staged in the
1967 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1968 * the block driver layer .bdrv_reopen_prepare()
1969 *
1970 * bs is the BlockDriverState to reopen
1971 * flags are the new open flags
1972 * queue is the reopen queue
1973 *
1974 * Returns 0 on success, non-zero on error. On error errp will be set
1975 * as well.
1976 *
1977 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1978 * It is the responsibility of the caller to then call the abort() or
1979 * commit() for any other BDS that have been left in a prepare() state
1980 *
1981 */
1982 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1983 Error **errp)
1984 {
1985 int ret = -1;
1986 Error *local_err = NULL;
1987 BlockDriver *drv;
1988 QemuOpts *opts;
1989 const char *value;
1990
1991 assert(reopen_state != NULL);
1992 assert(reopen_state->bs->drv != NULL);
1993 drv = reopen_state->bs->drv;
1994
1995 /* Process generic block layer options */
1996 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1997 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
1998 if (local_err) {
1999 error_propagate(errp, local_err);
2000 ret = -EINVAL;
2001 goto error;
2002 }
2003
2004 update_flags_from_options(&reopen_state->flags, opts);
2005
2006 /* node-name and driver must be unchanged. Put them back into the QDict, so
2007 * that they are checked at the end of this function. */
2008 value = qemu_opt_get(opts, "node-name");
2009 if (value) {
2010 qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2011 }
2012
2013 value = qemu_opt_get(opts, "driver");
2014 if (value) {
2015 qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2016 }
2017
2018 /* if we are to stay read-only, do not allow permission change
2019 * to r/w */
2020 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2021 reopen_state->flags & BDRV_O_RDWR) {
2022 error_setg(errp, "Node '%s' is read only",
2023 bdrv_get_device_or_node_name(reopen_state->bs));
2024 goto error;
2025 }
2026
2027
2028 ret = bdrv_flush(reopen_state->bs);
2029 if (ret) {
2030 error_setg_errno(errp, -ret, "Error flushing drive");
2031 goto error;
2032 }
2033
2034 if (drv->bdrv_reopen_prepare) {
2035 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2036 if (ret) {
2037 if (local_err != NULL) {
2038 error_propagate(errp, local_err);
2039 } else {
2040 error_setg(errp, "failed while preparing to reopen image '%s'",
2041 reopen_state->bs->filename);
2042 }
2043 goto error;
2044 }
2045 } else {
2046 /* It is currently mandatory to have a bdrv_reopen_prepare()
2047 * handler for each supported drv. */
2048 error_setg(errp, "Block format '%s' used by node '%s' "
2049 "does not support reopening files", drv->format_name,
2050 bdrv_get_device_or_node_name(reopen_state->bs));
2051 ret = -1;
2052 goto error;
2053 }
2054
2055 /* Options that are not handled are only okay if they are unchanged
2056 * compared to the old state. It is expected that some options are only
2057 * used for the initial open, but not reopen (e.g. filename) */
2058 if (qdict_size(reopen_state->options)) {
2059 const QDictEntry *entry = qdict_first(reopen_state->options);
2060
2061 do {
2062 QString *new_obj = qobject_to_qstring(entry->value);
2063 const char *new = qstring_get_str(new_obj);
2064 const char *old = qdict_get_try_str(reopen_state->bs->options,
2065 entry->key);
2066
2067 if (!old || strcmp(new, old)) {
2068 error_setg(errp, "Cannot change the option '%s'", entry->key);
2069 ret = -EINVAL;
2070 goto error;
2071 }
2072 } while ((entry = qdict_next(reopen_state->options, entry)));
2073 }
2074
2075 ret = 0;
2076
2077 error:
2078 qemu_opts_del(opts);
2079 return ret;
2080 }
2081
2082 /*
2083 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2084 * makes them final by swapping the staging BlockDriverState contents into
2085 * the active BlockDriverState contents.
2086 */
2087 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2088 {
2089 BlockDriver *drv;
2090
2091 assert(reopen_state != NULL);
2092 drv = reopen_state->bs->drv;
2093 assert(drv != NULL);
2094
2095 /* If there are any driver level actions to take */
2096 if (drv->bdrv_reopen_commit) {
2097 drv->bdrv_reopen_commit(reopen_state);
2098 }
2099
2100 /* set BDS specific flags now */
2101 QDECREF(reopen_state->bs->explicit_options);
2102
2103 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2104 reopen_state->bs->open_flags = reopen_state->flags;
2105 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2106
2107 bdrv_refresh_limits(reopen_state->bs, NULL);
2108 }
2109
2110 /*
2111 * Abort the reopen, and delete and free the staged changes in
2112 * reopen_state
2113 */
2114 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2115 {
2116 BlockDriver *drv;
2117
2118 assert(reopen_state != NULL);
2119 drv = reopen_state->bs->drv;
2120 assert(drv != NULL);
2121
2122 if (drv->bdrv_reopen_abort) {
2123 drv->bdrv_reopen_abort(reopen_state);
2124 }
2125
2126 QDECREF(reopen_state->explicit_options);
2127 }
2128
2129
2130 static void bdrv_close(BlockDriverState *bs)
2131 {
2132 BdrvAioNotifier *ban, *ban_next;
2133
2134 assert(!bs->job);
2135
2136 bdrv_drained_begin(bs); /* complete I/O */
2137 bdrv_flush(bs);
2138 bdrv_drain(bs); /* in case flush left pending I/O */
2139
2140 bdrv_release_named_dirty_bitmaps(bs);
2141 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2142
2143 bdrv_parent_cb_change_media(bs, false);
2144
2145 if (bs->drv) {
2146 BdrvChild *child, *next;
2147
2148 bs->drv->bdrv_close(bs);
2149 bs->drv = NULL;
2150
2151 bdrv_set_backing_hd(bs, NULL);
2152
2153 if (bs->file != NULL) {
2154 bdrv_unref_child(bs, bs->file);
2155 bs->file = NULL;
2156 }
2157
2158 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2159 /* TODO Remove bdrv_unref() from drivers' close function and use
2160 * bdrv_unref_child() here */
2161 if (child->bs->inherits_from == bs) {
2162 child->bs->inherits_from = NULL;
2163 }
2164 bdrv_detach_child(child);
2165 }
2166
2167 g_free(bs->opaque);
2168 bs->opaque = NULL;
2169 bs->copy_on_read = 0;
2170 bs->backing_file[0] = '\0';
2171 bs->backing_format[0] = '\0';
2172 bs->total_sectors = 0;
2173 bs->encrypted = 0;
2174 bs->valid_key = 0;
2175 bs->sg = 0;
2176 bs->zero_beyond_eof = false;
2177 QDECREF(bs->options);
2178 QDECREF(bs->explicit_options);
2179 bs->options = NULL;
2180 QDECREF(bs->full_open_options);
2181 bs->full_open_options = NULL;
2182 }
2183
2184 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2185 g_free(ban);
2186 }
2187 QLIST_INIT(&bs->aio_notifiers);
2188 bdrv_drained_end(bs);
2189 }
2190
2191 void bdrv_close_all(void)
2192 {
2193 BlockDriverState *bs;
2194 AioContext *aio_context;
2195
2196 /* Drop references from requests still in flight, such as canceled block
2197 * jobs whose AIO context has not been polled yet */
2198 bdrv_drain_all();
2199
2200 blk_remove_all_bs();
2201 blockdev_close_all_bdrv_states();
2202
2203 /* Cancel all block jobs */
2204 while (!QTAILQ_EMPTY(&all_bdrv_states)) {
2205 QTAILQ_FOREACH(bs, &all_bdrv_states, bs_list) {
2206 aio_context = bdrv_get_aio_context(bs);
2207
2208 aio_context_acquire(aio_context);
2209 if (bs->job) {
2210 block_job_cancel_sync(bs->job);
2211 aio_context_release(aio_context);
2212 break;
2213 }
2214 aio_context_release(aio_context);
2215 }
2216
2217 /* All the remaining BlockDriverStates are referenced directly or
2218 * indirectly from block jobs, so there needs to be at least one BDS
2219 * directly used by a block job */
2220 assert(bs);
2221 }
2222 }
2223
2224 static void change_parent_backing_link(BlockDriverState *from,
2225 BlockDriverState *to)
2226 {
2227 BdrvChild *c, *next;
2228
2229 if (from->blk) {
2230 /* FIXME We bypass blk_set_bs(), so we need to make these updates
2231 * manually. The root problem is not in this change function, but the
2232 * existence of BlockDriverState.blk. */
2233 to->blk = from->blk;
2234 from->blk = NULL;
2235 }
2236
2237 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
2238 assert(c->role != &child_backing);
2239 c->bs = to;
2240 QLIST_REMOVE(c, next_parent);
2241 QLIST_INSERT_HEAD(&to->parents, c, next_parent);
2242 bdrv_ref(to);
2243 bdrv_unref(from);
2244 }
2245 }
2246
2247 /*
2248 * Add new bs contents at the top of an image chain while the chain is
2249 * live, while keeping required fields on the top layer.
2250 *
2251 * This will modify the BlockDriverState fields, and swap contents
2252 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2253 *
2254 * bs_new must not be attached to a BlockBackend.
2255 *
2256 * This function does not create any image files.
2257 *
2258 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
2259 * that's what the callers commonly need. bs_new will be referenced by the old
2260 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
2261 * reference of its own, it must call bdrv_ref().
2262 */
2263 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2264 {
2265 assert(!bdrv_requests_pending(bs_top));
2266 assert(!bdrv_requests_pending(bs_new));
2267
2268 bdrv_ref(bs_top);
2269
2270 change_parent_backing_link(bs_top, bs_new);
2271 bdrv_set_backing_hd(bs_new, bs_top);
2272 bdrv_unref(bs_top);
2273
2274 /* bs_new is now referenced by its new parents, we don't need the
2275 * additional reference any more. */
2276 bdrv_unref(bs_new);
2277 }
2278
2279 void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
2280 {
2281 assert(!bdrv_requests_pending(old));
2282 assert(!bdrv_requests_pending(new));
2283
2284 bdrv_ref(old);
2285
2286 change_parent_backing_link(old, new);
2287
2288 /* Change backing files if a previously independent node is added to the
2289 * chain. For active commit, we replace top by its own (indirect) backing
2290 * file and don't do anything here so we don't build a loop. */
2291 if (new->backing == NULL && !bdrv_chain_contains(backing_bs(old), new)) {
2292 bdrv_set_backing_hd(new, backing_bs(old));
2293 bdrv_set_backing_hd(old, NULL);
2294 }
2295
2296 bdrv_unref(old);
2297 }
2298
2299 static void bdrv_delete(BlockDriverState *bs)
2300 {
2301 assert(!bs->job);
2302 assert(bdrv_op_blocker_is_empty(bs));
2303 assert(!bs->refcnt);
2304
2305 bdrv_close(bs);
2306
2307 /* remove from list, if necessary */
2308 if (bs->node_name[0] != '\0') {
2309 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
2310 }
2311 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
2312
2313 g_free(bs);
2314 }
2315
2316 /*
2317 * Run consistency checks on an image
2318 *
2319 * Returns 0 if the check could be completed (it doesn't mean that the image is
2320 * free of errors) or -errno when an internal error occurred. The results of the
2321 * check are stored in res.
2322 */
2323 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2324 {
2325 if (bs->drv == NULL) {
2326 return -ENOMEDIUM;
2327 }
2328 if (bs->drv->bdrv_check == NULL) {
2329 return -ENOTSUP;
2330 }
2331
2332 memset(res, 0, sizeof(*res));
2333 return bs->drv->bdrv_check(bs, res, fix);
2334 }
2335
2336 #define COMMIT_BUF_SECTORS 2048
2337
2338 /* commit COW file into the raw image */
2339 int bdrv_commit(BlockDriverState *bs)
2340 {
2341 BlockDriver *drv = bs->drv;
2342 int64_t sector, total_sectors, length, backing_length;
2343 int n, ro, open_flags;
2344 int ret = 0;
2345 uint8_t *buf = NULL;
2346
2347 if (!drv)
2348 return -ENOMEDIUM;
2349
2350 if (!bs->backing) {
2351 return -ENOTSUP;
2352 }
2353
2354 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, NULL) ||
2355 bdrv_op_is_blocked(bs->backing->bs, BLOCK_OP_TYPE_COMMIT_TARGET, NULL)) {
2356 return -EBUSY;
2357 }
2358
2359 ro = bs->backing->bs->read_only;
2360 open_flags = bs->backing->bs->open_flags;
2361
2362 if (ro) {
2363 if (bdrv_reopen(bs->backing->bs, open_flags | BDRV_O_RDWR, NULL)) {
2364 return -EACCES;
2365 }
2366 }
2367
2368 length = bdrv_getlength(bs);
2369 if (length < 0) {
2370 ret = length;
2371 goto ro_cleanup;
2372 }
2373
2374 backing_length = bdrv_getlength(bs->backing->bs);
2375 if (backing_length < 0) {
2376 ret = backing_length;
2377 goto ro_cleanup;
2378 }
2379
2380 /* If our top snapshot is larger than the backing file image,
2381 * grow the backing file image if possible. If not possible,
2382 * we must return an error */
2383 if (length > backing_length) {
2384 ret = bdrv_truncate(bs->backing->bs, length);
2385 if (ret < 0) {
2386 goto ro_cleanup;
2387 }
2388 }
2389
2390 total_sectors = length >> BDRV_SECTOR_BITS;
2391
2392 /* qemu_try_blockalign() for bs will choose an alignment that works for
2393 * bs->backing->bs as well, so no need to compare the alignment manually. */
2394 buf = qemu_try_blockalign(bs, COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2395 if (buf == NULL) {
2396 ret = -ENOMEM;
2397 goto ro_cleanup;
2398 }
2399
2400 for (sector = 0; sector < total_sectors; sector += n) {
2401 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2402 if (ret < 0) {
2403 goto ro_cleanup;
2404 }
2405 if (ret) {
2406 ret = bdrv_read(bs, sector, buf, n);
2407 if (ret < 0) {
2408 goto ro_cleanup;
2409 }
2410
2411 ret = bdrv_write(bs->backing->bs, sector, buf, n);
2412 if (ret < 0) {
2413 goto ro_cleanup;
2414 }
2415 }
2416 }
2417
2418 if (drv->bdrv_make_empty) {
2419 ret = drv->bdrv_make_empty(bs);
2420 if (ret < 0) {
2421 goto ro_cleanup;
2422 }
2423 bdrv_flush(bs);
2424 }
2425
2426 /*
2427 * Make sure all data we wrote to the backing device is actually
2428 * stable on disk.
2429 */
2430 if (bs->backing) {
2431 bdrv_flush(bs->backing->bs);
2432 }
2433
2434 ret = 0;
2435 ro_cleanup:
2436 qemu_vfree(buf);
2437
2438 if (ro) {
2439 /* ignoring error return here */
2440 bdrv_reopen(bs->backing->bs, open_flags & ~BDRV_O_RDWR, NULL);
2441 }
2442
2443 return ret;
2444 }
2445
2446 /*
2447 * Return values:
2448 * 0 - success
2449 * -EINVAL - backing format specified, but no file
2450 * -ENOSPC - can't update the backing file because no space is left in the
2451 * image file header
2452 * -ENOTSUP - format driver doesn't support changing the backing file
2453 */
2454 int bdrv_change_backing_file(BlockDriverState *bs,
2455 const char *backing_file, const char *backing_fmt)
2456 {
2457 BlockDriver *drv = bs->drv;
2458 int ret;
2459
2460 /* Backing file format doesn't make sense without a backing file */
2461 if (backing_fmt && !backing_file) {
2462 return -EINVAL;
2463 }
2464
2465 if (drv->bdrv_change_backing_file != NULL) {
2466 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2467 } else {
2468 ret = -ENOTSUP;
2469 }
2470
2471 if (ret == 0) {
2472 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2473 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2474 }
2475 return ret;
2476 }
2477
2478 /*
2479 * Finds the image layer in the chain that has 'bs' as its backing file.
2480 *
2481 * active is the current topmost image.
2482 *
2483 * Returns NULL if bs is not found in active's image chain,
2484 * or if active == bs.
2485 *
2486 * Returns the bottommost base image if bs == NULL.
2487 */
2488 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2489 BlockDriverState *bs)
2490 {
2491 while (active && bs != backing_bs(active)) {
2492 active = backing_bs(active);
2493 }
2494
2495 return active;
2496 }
2497
2498 /* Given a BDS, searches for the base layer. */
2499 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
2500 {
2501 return bdrv_find_overlay(bs, NULL);
2502 }
2503
2504 /*
2505 * Drops images above 'base' up to and including 'top', and sets the image
2506 * above 'top' to have base as its backing file.
2507 *
2508 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2509 * information in 'bs' can be properly updated.
2510 *
2511 * E.g., this will convert the following chain:
2512 * bottom <- base <- intermediate <- top <- active
2513 *
2514 * to
2515 *
2516 * bottom <- base <- active
2517 *
2518 * It is allowed for bottom==base, in which case it converts:
2519 *
2520 * base <- intermediate <- top <- active
2521 *
2522 * to
2523 *
2524 * base <- active
2525 *
2526 * If backing_file_str is non-NULL, it will be used when modifying top's
2527 * overlay image metadata.
2528 *
2529 * Error conditions:
2530 * if active == top, that is considered an error
2531 *
2532 */
2533 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2534 BlockDriverState *base, const char *backing_file_str)
2535 {
2536 BlockDriverState *new_top_bs = NULL;
2537 int ret = -EIO;
2538
2539 if (!top->drv || !base->drv) {
2540 goto exit;
2541 }
2542
2543 new_top_bs = bdrv_find_overlay(active, top);
2544
2545 if (new_top_bs == NULL) {
2546 /* we could not find the image above 'top', this is an error */
2547 goto exit;
2548 }
2549
2550 /* special case of new_top_bs->backing->bs already pointing to base - nothing
2551 * to do, no intermediate images */
2552 if (backing_bs(new_top_bs) == base) {
2553 ret = 0;
2554 goto exit;
2555 }
2556
2557 /* Make sure that base is in the backing chain of top */
2558 if (!bdrv_chain_contains(top, base)) {
2559 goto exit;
2560 }
2561
2562 /* success - we can delete the intermediate states, and link top->base */
2563 backing_file_str = backing_file_str ? backing_file_str : base->filename;
2564 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
2565 base->drv ? base->drv->format_name : "");
2566 if (ret) {
2567 goto exit;
2568 }
2569 bdrv_set_backing_hd(new_top_bs, base);
2570
2571 ret = 0;
2572 exit:
2573 return ret;
2574 }
2575
2576 /**
2577 * Truncate file to 'offset' bytes (needed only for file protocols)
2578 */
2579 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2580 {
2581 BlockDriver *drv = bs->drv;
2582 int ret;
2583 if (!drv)
2584 return -ENOMEDIUM;
2585 if (!drv->bdrv_truncate)
2586 return -ENOTSUP;
2587 if (bs->read_only)
2588 return -EACCES;
2589
2590 ret = drv->bdrv_truncate(bs, offset);
2591 if (ret == 0) {
2592 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2593 bdrv_dirty_bitmap_truncate(bs);
2594 bdrv_parent_cb_resize(bs);
2595 }
2596 return ret;
2597 }
2598
2599 /**
2600 * Length of a allocated file in bytes. Sparse files are counted by actual
2601 * allocated space. Return < 0 if error or unknown.
2602 */
2603 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2604 {
2605 BlockDriver *drv = bs->drv;
2606 if (!drv) {
2607 return -ENOMEDIUM;
2608 }
2609 if (drv->bdrv_get_allocated_file_size) {
2610 return drv->bdrv_get_allocated_file_size(bs);
2611 }
2612 if (bs->file) {
2613 return bdrv_get_allocated_file_size(bs->file->bs);
2614 }
2615 return -ENOTSUP;
2616 }
2617
2618 /**
2619 * Return number of sectors on success, -errno on error.
2620 */
2621 int64_t bdrv_nb_sectors(BlockDriverState *bs)
2622 {
2623 BlockDriver *drv = bs->drv;
2624
2625 if (!drv)
2626 return -ENOMEDIUM;
2627
2628 if (drv->has_variable_length) {
2629 int ret = refresh_total_sectors(bs, bs->total_sectors);
2630 if (ret < 0) {
2631 return ret;
2632 }
2633 }
2634 return bs->total_sectors;
2635 }
2636
2637 /**
2638 * Return length in bytes on success, -errno on error.
2639 * The length is always a multiple of BDRV_SECTOR_SIZE.
2640 */
2641 int64_t bdrv_getlength(BlockDriverState *bs)
2642 {
2643 int64_t ret = bdrv_nb_sectors(bs);
2644
2645 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
2646 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
2647 }
2648
2649 /* return 0 as number of sectors if no device present or error */
2650 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2651 {
2652 int64_t nb_sectors = bdrv_nb_sectors(bs);
2653
2654 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
2655 }
2656
2657 int bdrv_is_read_only(BlockDriverState *bs)
2658 {
2659 return bs->read_only;
2660 }
2661
2662 int bdrv_is_sg(BlockDriverState *bs)
2663 {
2664 return bs->sg;
2665 }
2666
2667 int bdrv_is_encrypted(BlockDriverState *bs)
2668 {
2669 if (bs->backing && bs->backing->bs->encrypted) {
2670 return 1;
2671 }
2672 return bs->encrypted;
2673 }
2674
2675 int bdrv_key_required(BlockDriverState *bs)
2676 {
2677 BdrvChild *backing = bs->backing;
2678
2679 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
2680 return 1;
2681 }
2682 return (bs->encrypted && !bs->valid_key);
2683 }
2684
2685 int bdrv_set_key(BlockDriverState *bs, const char *key)
2686 {
2687 int ret;
2688 if (bs->backing && bs->backing->bs->encrypted) {
2689 ret = bdrv_set_key(bs->backing->bs, key);
2690 if (ret < 0)
2691 return ret;
2692 if (!bs->encrypted)
2693 return 0;
2694 }
2695 if (!bs->encrypted) {
2696 return -EINVAL;
2697 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2698 return -ENOMEDIUM;
2699 }
2700 ret = bs->drv->bdrv_set_key(bs, key);
2701 if (ret < 0) {
2702 bs->valid_key = 0;
2703 } else if (!bs->valid_key) {
2704 /* call the change callback now, we skipped it on open */
2705 bs->valid_key = 1;
2706 bdrv_parent_cb_change_media(bs, true);
2707 }
2708 return ret;
2709 }
2710
2711 /*
2712 * Provide an encryption key for @bs.
2713 * If @key is non-null:
2714 * If @bs is not encrypted, fail.
2715 * Else if the key is invalid, fail.
2716 * Else set @bs's key to @key, replacing the existing key, if any.
2717 * If @key is null:
2718 * If @bs is encrypted and still lacks a key, fail.
2719 * Else do nothing.
2720 * On failure, store an error object through @errp if non-null.
2721 */
2722 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
2723 {
2724 if (key) {
2725 if (!bdrv_is_encrypted(bs)) {
2726 error_setg(errp, "Node '%s' is not encrypted",
2727 bdrv_get_device_or_node_name(bs));
2728 } else if (bdrv_set_key(bs, key) < 0) {
2729 error_setg(errp, QERR_INVALID_PASSWORD);
2730 }
2731 } else {
2732 if (bdrv_key_required(bs)) {
2733 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
2734 "'%s' (%s) is encrypted",
2735 bdrv_get_device_or_node_name(bs),
2736 bdrv_get_encrypted_filename(bs));
2737 }
2738 }
2739 }
2740
2741 const char *bdrv_get_format_name(BlockDriverState *bs)
2742 {
2743 return bs->drv ? bs->drv->format_name : NULL;
2744 }
2745
2746 static int qsort_strcmp(const void *a, const void *b)
2747 {
2748 return strcmp(a, b);
2749 }
2750
2751 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2752 void *opaque)
2753 {
2754 BlockDriver *drv;
2755 int count = 0;
2756 int i;
2757 const char **formats = NULL;
2758
2759 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2760 if (drv->format_name) {
2761 bool found = false;
2762 int i = count;
2763 while (formats && i && !found) {
2764 found = !strcmp(formats[--i], drv->format_name);
2765 }
2766
2767 if (!found) {
2768 formats = g_renew(const char *, formats, count + 1);
2769 formats[count++] = drv->format_name;
2770 }
2771 }
2772 }
2773
2774 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
2775
2776 for (i = 0; i < count; i++) {
2777 it(opaque, formats[i]);
2778 }
2779
2780 g_free(formats);
2781 }
2782
2783 /* This function is to find a node in the bs graph */
2784 BlockDriverState *bdrv_find_node(const char *node_name)
2785 {
2786 BlockDriverState *bs;
2787
2788 assert(node_name);
2789
2790 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2791 if (!strcmp(node_name, bs->node_name)) {
2792 return bs;
2793 }
2794 }
2795 return NULL;
2796 }
2797
2798 /* Put this QMP function here so it can access the static graph_bdrv_states. */
2799 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
2800 {
2801 BlockDeviceInfoList *list, *entry;
2802 BlockDriverState *bs;
2803
2804 list = NULL;
2805 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2806 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
2807 if (!info) {
2808 qapi_free_BlockDeviceInfoList(list);
2809 return NULL;
2810 }
2811 entry = g_malloc0(sizeof(*entry));
2812 entry->value = info;
2813 entry->next = list;
2814 list = entry;
2815 }
2816
2817 return list;
2818 }
2819
2820 BlockDriverState *bdrv_lookup_bs(const char *device,
2821 const char *node_name,
2822 Error **errp)
2823 {
2824 BlockBackend *blk;
2825 BlockDriverState *bs;
2826
2827 if (device) {
2828 blk = blk_by_name(device);
2829
2830 if (blk) {
2831 bs = blk_bs(blk);
2832 if (!bs) {
2833 error_setg(errp, "Device '%s' has no medium", device);
2834 }
2835
2836 return bs;
2837 }
2838 }
2839
2840 if (node_name) {
2841 bs = bdrv_find_node(node_name);
2842
2843 if (bs) {
2844 return bs;
2845 }
2846 }
2847
2848 error_setg(errp, "Cannot find device=%s nor node_name=%s",
2849 device ? device : "",
2850 node_name ? node_name : "");
2851 return NULL;
2852 }
2853
2854 /* If 'base' is in the same chain as 'top', return true. Otherwise,
2855 * return false. If either argument is NULL, return false. */
2856 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
2857 {
2858 while (top && top != base) {
2859 top = backing_bs(top);
2860 }
2861
2862 return top != NULL;
2863 }
2864
2865 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
2866 {
2867 if (!bs) {
2868 return QTAILQ_FIRST(&graph_bdrv_states);
2869 }
2870 return QTAILQ_NEXT(bs, node_list);
2871 }
2872
2873 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
2874 * the monitor or attached to a BlockBackend */
2875 BlockDriverState *bdrv_next(BlockDriverState *bs)
2876 {
2877 if (!bs || bs->blk) {
2878 bs = blk_next_root_bs(bs);
2879 if (bs) {
2880 return bs;
2881 }
2882 }
2883
2884 /* Ignore all BDSs that are attached to a BlockBackend here; they have been
2885 * handled by the above block already */
2886 do {
2887 bs = bdrv_next_monitor_owned(bs);
2888 } while (bs && bs->blk);
2889 return bs;
2890 }
2891
2892 const char *bdrv_get_node_name(const BlockDriverState *bs)
2893 {
2894 return bs->node_name;
2895 }
2896
2897 /* TODO check what callers really want: bs->node_name or blk_name() */
2898 const char *bdrv_get_device_name(const BlockDriverState *bs)
2899 {
2900 return bs->blk ? blk_name(bs->blk) : "";
2901 }
2902
2903 /* This can be used to identify nodes that might not have a device
2904 * name associated. Since node and device names live in the same
2905 * namespace, the result is unambiguous. The exception is if both are
2906 * absent, then this returns an empty (non-null) string. */
2907 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
2908 {
2909 return bs->blk ? blk_name(bs->blk) : bs->node_name;
2910 }
2911
2912 int bdrv_get_flags(BlockDriverState *bs)
2913 {
2914 return bs->open_flags;
2915 }
2916
2917 int bdrv_has_zero_init_1(BlockDriverState *bs)
2918 {
2919 return 1;
2920 }
2921
2922 int bdrv_has_zero_init(BlockDriverState *bs)
2923 {
2924 assert(bs->drv);
2925
2926 /* If BS is a copy on write image, it is initialized to
2927 the contents of the base image, which may not be zeroes. */
2928 if (bs->backing) {
2929 return 0;
2930 }
2931 if (bs->drv->bdrv_has_zero_init) {
2932 return bs->drv->bdrv_has_zero_init(bs);
2933 }
2934
2935 /* safe default */
2936 return 0;
2937 }
2938
2939 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
2940 {
2941 BlockDriverInfo bdi;
2942
2943 if (bs->backing) {
2944 return false;
2945 }
2946
2947 if (bdrv_get_info(bs, &bdi) == 0) {
2948 return bdi.unallocated_blocks_are_zero;
2949 }
2950
2951 return false;
2952 }
2953
2954 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
2955 {
2956 BlockDriverInfo bdi;
2957
2958 if (bs->backing || !(bs->open_flags & BDRV_O_UNMAP)) {
2959 return false;
2960 }
2961
2962 if (bdrv_get_info(bs, &bdi) == 0) {
2963 return bdi.can_write_zeroes_with_unmap;
2964 }
2965
2966 return false;
2967 }
2968
2969 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2970 {
2971 if (bs->backing && bs->backing->bs->encrypted)
2972 return bs->backing_file;
2973 else if (bs->encrypted)
2974 return bs->filename;
2975 else
2976 return NULL;
2977 }
2978
2979 void bdrv_get_backing_filename(BlockDriverState *bs,
2980 char *filename, int filename_size)
2981 {
2982 pstrcpy(filename, filename_size, bs->backing_file);
2983 }
2984
2985 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2986 {
2987 BlockDriver *drv = bs->drv;
2988 if (!drv)
2989 return -ENOMEDIUM;
2990 if (!drv->bdrv_get_info)
2991 return -ENOTSUP;
2992 memset(bdi, 0, sizeof(*bdi));
2993 return drv->bdrv_get_info(bs, bdi);
2994 }
2995
2996 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
2997 {
2998 BlockDriver *drv = bs->drv;
2999 if (drv && drv->bdrv_get_specific_info) {
3000 return drv->bdrv_get_specific_info(bs);
3001 }
3002 return NULL;
3003 }
3004
3005 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3006 {
3007 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3008 return;
3009 }
3010
3011 bs->drv->bdrv_debug_event(bs, event);
3012 }
3013
3014 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3015 const char *tag)
3016 {
3017 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3018 bs = bs->file ? bs->file->bs : NULL;
3019 }
3020
3021 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3022 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3023 }
3024
3025 return -ENOTSUP;
3026 }
3027
3028 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3029 {
3030 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3031 bs = bs->file ? bs->file->bs : NULL;
3032 }
3033
3034 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3035 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3036 }
3037
3038 return -ENOTSUP;
3039 }
3040
3041 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3042 {
3043 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3044 bs = bs->file ? bs->file->bs : NULL;
3045 }
3046
3047 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3048 return bs->drv->bdrv_debug_resume(bs, tag);
3049 }
3050
3051 return -ENOTSUP;
3052 }
3053
3054 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3055 {
3056 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3057 bs = bs->file ? bs->file->bs : NULL;
3058 }
3059
3060 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3061 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3062 }
3063
3064 return false;
3065 }
3066
3067 int bdrv_is_snapshot(BlockDriverState *bs)
3068 {
3069 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
3070 }
3071
3072 /* backing_file can either be relative, or absolute, or a protocol. If it is
3073 * relative, it must be relative to the chain. So, passing in bs->filename
3074 * from a BDS as backing_file should not be done, as that may be relative to
3075 * the CWD rather than the chain. */
3076 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3077 const char *backing_file)
3078 {
3079 char *filename_full = NULL;
3080 char *backing_file_full = NULL;
3081 char *filename_tmp = NULL;
3082 int is_protocol = 0;
3083 BlockDriverState *curr_bs = NULL;
3084 BlockDriverState *retval = NULL;
3085
3086 if (!bs || !bs->drv || !backing_file) {
3087 return NULL;
3088 }
3089
3090 filename_full = g_malloc(PATH_MAX);
3091 backing_file_full = g_malloc(PATH_MAX);
3092 filename_tmp = g_malloc(PATH_MAX);
3093
3094 is_protocol = path_has_protocol(backing_file);
3095
3096 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3097
3098 /* If either of the filename paths is actually a protocol, then
3099 * compare unmodified paths; otherwise make paths relative */
3100 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3101 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3102 retval = curr_bs->backing->bs;
3103 break;
3104 }
3105 } else {
3106 /* If not an absolute filename path, make it relative to the current
3107 * image's filename path */
3108 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3109 backing_file);
3110
3111 /* We are going to compare absolute pathnames */
3112 if (!realpath(filename_tmp, filename_full)) {
3113 continue;
3114 }
3115
3116 /* We need to make sure the backing filename we are comparing against
3117 * is relative to the current image filename (or absolute) */
3118 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3119 curr_bs->backing_file);
3120
3121 if (!realpath(filename_tmp, backing_file_full)) {
3122 continue;
3123 }
3124
3125 if (strcmp(backing_file_full, filename_full) == 0) {
3126 retval = curr_bs->backing->bs;
3127 break;
3128 }
3129 }
3130 }
3131
3132 g_free(filename_full);
3133 g_free(backing_file_full);
3134 g_free(filename_tmp);
3135 return retval;
3136 }
3137
3138 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3139 {
3140 if (!bs->drv) {
3141 return 0;
3142 }
3143
3144 if (!bs->backing) {
3145 return 0;
3146 }
3147
3148 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3149 }
3150
3151 void bdrv_init(void)
3152 {
3153 module_call_init(MODULE_INIT_BLOCK);
3154 }
3155
3156 void bdrv_init_with_whitelist(void)
3157 {
3158 use_bdrv_whitelist = 1;
3159 bdrv_init();
3160 }
3161
3162 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3163 {
3164 BdrvChild *child;
3165 Error *local_err = NULL;
3166 int ret;
3167
3168 if (!bs->drv) {
3169 return;
3170 }
3171
3172 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3173 return;
3174 }
3175 bs->open_flags &= ~BDRV_O_INACTIVE;
3176
3177 if (bs->drv->bdrv_invalidate_cache) {
3178 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3179 if (local_err) {
3180 bs->open_flags |= BDRV_O_INACTIVE;
3181 error_propagate(errp, local_err);
3182 return;
3183 }
3184 }
3185
3186 QLIST_FOREACH(child, &bs->children, next) {
3187 bdrv_invalidate_cache(child->bs, &local_err);
3188 if (local_err) {
3189 bs->open_flags |= BDRV_O_INACTIVE;
3190 error_propagate(errp, local_err);
3191 return;
3192 }
3193 }
3194
3195 ret = refresh_total_sectors(bs, bs->total_sectors);
3196 if (ret < 0) {
3197 bs->open_flags |= BDRV_O_INACTIVE;
3198 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3199 return;
3200 }
3201 }
3202
3203 void bdrv_invalidate_cache_all(Error **errp)
3204 {
3205 BlockDriverState *bs = NULL;
3206 Error *local_err = NULL;
3207
3208 while ((bs = bdrv_next(bs)) != NULL) {
3209 AioContext *aio_context = bdrv_get_aio_context(bs);
3210
3211 aio_context_acquire(aio_context);
3212 bdrv_invalidate_cache(bs, &local_err);
3213 aio_context_release(aio_context);
3214 if (local_err) {
3215 error_propagate(errp, local_err);
3216 return;
3217 }
3218 }
3219 }
3220
3221 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3222 bool setting_flag)
3223 {
3224 BdrvChild *child;
3225 int ret;
3226
3227 if (!setting_flag && bs->drv->bdrv_inactivate) {
3228 ret = bs->drv->bdrv_inactivate(bs);
3229 if (ret < 0) {
3230 return ret;
3231 }
3232 }
3233
3234 QLIST_FOREACH(child, &bs->children, next) {
3235 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
3236 if (ret < 0) {
3237 return ret;
3238 }
3239 }
3240
3241 if (setting_flag) {
3242 bs->open_flags |= BDRV_O_INACTIVE;
3243 }
3244 return 0;
3245 }
3246
3247 int bdrv_inactivate_all(void)
3248 {
3249 BlockDriverState *bs = NULL;
3250 int ret = 0;
3251 int pass;
3252
3253 while ((bs = bdrv_next(bs)) != NULL) {
3254 aio_context_acquire(bdrv_get_aio_context(bs));
3255 }
3256
3257 /* We do two passes of inactivation. The first pass calls to drivers'
3258 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
3259 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
3260 * is allowed. */
3261 for (pass = 0; pass < 2; pass++) {
3262 bs = NULL;
3263 while ((bs = bdrv_next(bs)) != NULL) {
3264 ret = bdrv_inactivate_recurse(bs, pass);
3265 if (ret < 0) {
3266 goto out;
3267 }
3268 }
3269 }
3270
3271 out:
3272 bs = NULL;
3273 while ((bs = bdrv_next(bs)) != NULL) {
3274 aio_context_release(bdrv_get_aio_context(bs));
3275 }
3276
3277 return ret;
3278 }
3279
3280 /**************************************************************/
3281 /* removable device support */
3282
3283 /**
3284 * Return TRUE if the media is present
3285 */
3286 bool bdrv_is_inserted(BlockDriverState *bs)
3287 {
3288 BlockDriver *drv = bs->drv;
3289 BdrvChild *child;
3290
3291 if (!drv) {
3292 return false;
3293 }
3294 if (drv->bdrv_is_inserted) {
3295 return drv->bdrv_is_inserted(bs);
3296 }
3297 QLIST_FOREACH(child, &bs->children, next) {
3298 if (!bdrv_is_inserted(child->bs)) {
3299 return false;
3300 }
3301 }
3302 return true;
3303 }
3304
3305 /**
3306 * Return whether the media changed since the last call to this
3307 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3308 */
3309 int bdrv_media_changed(BlockDriverState *bs)
3310 {
3311 BlockDriver *drv = bs->drv;
3312
3313 if (drv && drv->bdrv_media_changed) {
3314 return drv->bdrv_media_changed(bs);
3315 }
3316 return -ENOTSUP;
3317 }
3318
3319 /**
3320 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3321 */
3322 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3323 {
3324 BlockDriver *drv = bs->drv;
3325 const char *device_name;
3326
3327 if (drv && drv->bdrv_eject) {
3328 drv->bdrv_eject(bs, eject_flag);
3329 }
3330
3331 device_name = bdrv_get_device_name(bs);
3332 if (device_name[0] != '\0') {
3333 qapi_event_send_device_tray_moved(device_name,
3334 eject_flag, &error_abort);
3335 }
3336 }
3337
3338 /**
3339 * Lock or unlock the media (if it is locked, the user won't be able
3340 * to eject it manually).
3341 */
3342 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3343 {
3344 BlockDriver *drv = bs->drv;
3345
3346 trace_bdrv_lock_medium(bs, locked);
3347
3348 if (drv && drv->bdrv_lock_medium) {
3349 drv->bdrv_lock_medium(bs, locked);
3350 }
3351 }
3352
3353 /* Get a reference to bs */
3354 void bdrv_ref(BlockDriverState *bs)
3355 {
3356 bs->refcnt++;
3357 }
3358
3359 /* Release a previously grabbed reference to bs.
3360 * If after releasing, reference count is zero, the BlockDriverState is
3361 * deleted. */
3362 void bdrv_unref(BlockDriverState *bs)
3363 {
3364 if (!bs) {
3365 return;
3366 }
3367 assert(bs->refcnt > 0);
3368 if (--bs->refcnt == 0) {
3369 bdrv_delete(bs);
3370 }
3371 }
3372
3373 struct BdrvOpBlocker {
3374 Error *reason;
3375 QLIST_ENTRY(BdrvOpBlocker) list;
3376 };
3377
3378 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
3379 {
3380 BdrvOpBlocker *blocker;
3381 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3382 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
3383 blocker = QLIST_FIRST(&bs->op_blockers[op]);
3384 if (errp) {
3385 *errp = error_copy(blocker->reason);
3386 error_prepend(errp, "Node '%s' is busy: ",
3387 bdrv_get_device_or_node_name(bs));
3388 }
3389 return true;
3390 }
3391 return false;
3392 }
3393
3394 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
3395 {
3396 BdrvOpBlocker *blocker;
3397 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3398
3399 blocker = g_new0(BdrvOpBlocker, 1);
3400 blocker->reason = reason;
3401 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
3402 }
3403
3404 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
3405 {
3406 BdrvOpBlocker *blocker, *next;
3407 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3408 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
3409 if (blocker->reason == reason) {
3410 QLIST_REMOVE(blocker, list);
3411 g_free(blocker);
3412 }
3413 }
3414 }
3415
3416 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
3417 {
3418 int i;
3419 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3420 bdrv_op_block(bs, i, reason);
3421 }
3422 }
3423
3424 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
3425 {
3426 int i;
3427 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3428 bdrv_op_unblock(bs, i, reason);
3429 }
3430 }
3431
3432 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
3433 {
3434 int i;
3435
3436 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3437 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
3438 return false;
3439 }
3440 }
3441 return true;
3442 }
3443
3444 void bdrv_img_create(const char *filename, const char *fmt,
3445 const char *base_filename, const char *base_fmt,
3446 char *options, uint64_t img_size, int flags,
3447 Error **errp, bool quiet)
3448 {
3449 QemuOptsList *create_opts = NULL;
3450 QemuOpts *opts = NULL;
3451 const char *backing_fmt, *backing_file;
3452 int64_t size;
3453 BlockDriver *drv, *proto_drv;
3454 Error *local_err = NULL;
3455 int ret = 0;
3456
3457 /* Find driver and parse its options */
3458 drv = bdrv_find_format(fmt);
3459 if (!drv) {
3460 error_setg(errp, "Unknown file format '%s'", fmt);
3461 return;
3462 }
3463
3464 proto_drv = bdrv_find_protocol(filename, true, errp);
3465 if (!proto_drv) {
3466 return;
3467 }
3468
3469 if (!drv->create_opts) {
3470 error_setg(errp, "Format driver '%s' does not support image creation",
3471 drv->format_name);
3472 return;
3473 }
3474
3475 if (!proto_drv->create_opts) {
3476 error_setg(errp, "Protocol driver '%s' does not support image creation",
3477 proto_drv->format_name);
3478 return;
3479 }
3480
3481 create_opts = qemu_opts_append(create_opts, drv->create_opts);
3482 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
3483
3484 /* Create parameter list with default values */
3485 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
3486 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
3487
3488 /* Parse -o options */
3489 if (options) {
3490 qemu_opts_do_parse(opts, options, NULL, &local_err);
3491 if (local_err) {
3492 error_report_err(local_err);
3493 local_err = NULL;
3494 error_setg(errp, "Invalid options for file format '%s'", fmt);
3495 goto out;
3496 }
3497 }
3498
3499 if (base_filename) {
3500 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
3501 if (local_err) {
3502 error_setg(errp, "Backing file not supported for file format '%s'",
3503 fmt);
3504 goto out;
3505 }
3506 }
3507
3508 if (base_fmt) {
3509 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
3510 if (local_err) {
3511 error_setg(errp, "Backing file format not supported for file "
3512 "format '%s'", fmt);
3513 goto out;
3514 }
3515 }
3516
3517 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3518 if (backing_file) {
3519 if (!strcmp(filename, backing_file)) {
3520 error_setg(errp, "Error: Trying to create an image with the "
3521 "same filename as the backing file");
3522 goto out;
3523 }
3524 }
3525
3526 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3527
3528 // The size for the image must always be specified, with one exception:
3529 // If we are using a backing file, we can obtain the size from there
3530 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3531 if (size == -1) {
3532 if (backing_file) {
3533 BlockDriverState *bs;
3534 char *full_backing = g_new0(char, PATH_MAX);
3535 int64_t size;
3536 int back_flags;
3537 QDict *backing_options = NULL;
3538
3539 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
3540 full_backing, PATH_MAX,
3541 &local_err);
3542 if (local_err) {
3543 g_free(full_backing);
3544 goto out;
3545 }
3546
3547 /* backing files always opened read-only */
3548 back_flags = flags;
3549 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
3550
3551 if (backing_fmt) {
3552 backing_options = qdict_new();
3553 qdict_put(backing_options, "driver",
3554 qstring_from_str(backing_fmt));
3555 }
3556
3557 bs = NULL;
3558 ret = bdrv_open(&bs, full_backing, NULL, backing_options,
3559 back_flags, &local_err);
3560 g_free(full_backing);
3561 if (ret < 0) {
3562 goto out;
3563 }
3564 size = bdrv_getlength(bs);
3565 if (size < 0) {
3566 error_setg_errno(errp, -size, "Could not get size of '%s'",
3567 backing_file);
3568 bdrv_unref(bs);
3569 goto out;
3570 }
3571
3572 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
3573
3574 bdrv_unref(bs);
3575 } else {
3576 error_setg(errp, "Image creation needs a size parameter");
3577 goto out;
3578 }
3579 }
3580
3581 if (!quiet) {
3582 printf("Formatting '%s', fmt=%s ", filename, fmt);