nbd: Don't use *_to_cpup() functions
[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 BlockDriverState *bdrv_open_inherit(const char *filename,
68 const char *reference,
69 QDict *options, int flags,
70 BlockDriverState *parent,
71 const BdrvChildRole *child_role,
72 Error **errp);
73
74 /* If non-zero, use only whitelisted block drivers */
75 static int use_bdrv_whitelist;
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(void)
224 {
225 BlockDriverState *bs;
226 int i;
227
228 bs = g_new0(BlockDriverState, 1);
229 QLIST_INIT(&bs->dirty_bitmaps);
230 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
231 QLIST_INIT(&bs->op_blockers[i]);
232 }
233 notifier_with_return_list_init(&bs->before_write_notifiers);
234 bs->refcnt = 1;
235 bs->aio_context = qemu_get_aio_context();
236
237 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
238
239 return bs;
240 }
241
242 BlockDriver *bdrv_find_format(const char *format_name)
243 {
244 BlockDriver *drv1;
245 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
246 if (!strcmp(drv1->format_name, format_name)) {
247 return drv1;
248 }
249 }
250 return NULL;
251 }
252
253 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
254 {
255 static const char *whitelist_rw[] = {
256 CONFIG_BDRV_RW_WHITELIST
257 };
258 static const char *whitelist_ro[] = {
259 CONFIG_BDRV_RO_WHITELIST
260 };
261 const char **p;
262
263 if (!whitelist_rw[0] && !whitelist_ro[0]) {
264 return 1; /* no whitelist, anything goes */
265 }
266
267 for (p = whitelist_rw; *p; p++) {
268 if (!strcmp(drv->format_name, *p)) {
269 return 1;
270 }
271 }
272 if (read_only) {
273 for (p = whitelist_ro; *p; p++) {
274 if (!strcmp(drv->format_name, *p)) {
275 return 1;
276 }
277 }
278 }
279 return 0;
280 }
281
282 bool bdrv_uses_whitelist(void)
283 {
284 return use_bdrv_whitelist;
285 }
286
287 typedef struct CreateCo {
288 BlockDriver *drv;
289 char *filename;
290 QemuOpts *opts;
291 int ret;
292 Error *err;
293 } CreateCo;
294
295 static void coroutine_fn bdrv_create_co_entry(void *opaque)
296 {
297 Error *local_err = NULL;
298 int ret;
299
300 CreateCo *cco = opaque;
301 assert(cco->drv);
302
303 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
304 if (local_err) {
305 error_propagate(&cco->err, local_err);
306 }
307 cco->ret = ret;
308 }
309
310 int bdrv_create(BlockDriver *drv, const char* filename,
311 QemuOpts *opts, Error **errp)
312 {
313 int ret;
314
315 Coroutine *co;
316 CreateCo cco = {
317 .drv = drv,
318 .filename = g_strdup(filename),
319 .opts = opts,
320 .ret = NOT_DONE,
321 .err = NULL,
322 };
323
324 if (!drv->bdrv_create) {
325 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
326 ret = -ENOTSUP;
327 goto out;
328 }
329
330 if (qemu_in_coroutine()) {
331 /* Fast-path if already in coroutine context */
332 bdrv_create_co_entry(&cco);
333 } else {
334 co = qemu_coroutine_create(bdrv_create_co_entry);
335 qemu_coroutine_enter(co, &cco);
336 while (cco.ret == NOT_DONE) {
337 aio_poll(qemu_get_aio_context(), true);
338 }
339 }
340
341 ret = cco.ret;
342 if (ret < 0) {
343 if (cco.err) {
344 error_propagate(errp, cco.err);
345 } else {
346 error_setg_errno(errp, -ret, "Could not create image");
347 }
348 }
349
350 out:
351 g_free(cco.filename);
352 return ret;
353 }
354
355 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
356 {
357 BlockDriver *drv;
358 Error *local_err = NULL;
359 int ret;
360
361 drv = bdrv_find_protocol(filename, true, errp);
362 if (drv == NULL) {
363 return -ENOENT;
364 }
365
366 ret = bdrv_create(drv, filename, opts, &local_err);
367 if (local_err) {
368 error_propagate(errp, local_err);
369 }
370 return ret;
371 }
372
373 /**
374 * Try to get @bs's logical and physical block size.
375 * On success, store them in @bsz struct and return 0.
376 * On failure return -errno.
377 * @bs must not be empty.
378 */
379 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
380 {
381 BlockDriver *drv = bs->drv;
382
383 if (drv && drv->bdrv_probe_blocksizes) {
384 return drv->bdrv_probe_blocksizes(bs, bsz);
385 }
386
387 return -ENOTSUP;
388 }
389
390 /**
391 * Try to get @bs's geometry (cyls, heads, sectors).
392 * On success, store them in @geo struct and return 0.
393 * On failure return -errno.
394 * @bs must not be empty.
395 */
396 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
397 {
398 BlockDriver *drv = bs->drv;
399
400 if (drv && drv->bdrv_probe_geometry) {
401 return drv->bdrv_probe_geometry(bs, geo);
402 }
403
404 return -ENOTSUP;
405 }
406
407 /*
408 * Create a uniquely-named empty temporary file.
409 * Return 0 upon success, otherwise a negative errno value.
410 */
411 int get_tmp_filename(char *filename, int size)
412 {
413 #ifdef _WIN32
414 char temp_dir[MAX_PATH];
415 /* GetTempFileName requires that its output buffer (4th param)
416 have length MAX_PATH or greater. */
417 assert(size >= MAX_PATH);
418 return (GetTempPath(MAX_PATH, temp_dir)
419 && GetTempFileName(temp_dir, "qem", 0, filename)
420 ? 0 : -GetLastError());
421 #else
422 int fd;
423 const char *tmpdir;
424 tmpdir = getenv("TMPDIR");
425 if (!tmpdir) {
426 tmpdir = "/var/tmp";
427 }
428 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
429 return -EOVERFLOW;
430 }
431 fd = mkstemp(filename);
432 if (fd < 0) {
433 return -errno;
434 }
435 if (close(fd) != 0) {
436 unlink(filename);
437 return -errno;
438 }
439 return 0;
440 #endif
441 }
442
443 /*
444 * Detect host devices. By convention, /dev/cdrom[N] is always
445 * recognized as a host CDROM.
446 */
447 static BlockDriver *find_hdev_driver(const char *filename)
448 {
449 int score_max = 0, score;
450 BlockDriver *drv = NULL, *d;
451
452 QLIST_FOREACH(d, &bdrv_drivers, list) {
453 if (d->bdrv_probe_device) {
454 score = d->bdrv_probe_device(filename);
455 if (score > score_max) {
456 score_max = score;
457 drv = d;
458 }
459 }
460 }
461
462 return drv;
463 }
464
465 BlockDriver *bdrv_find_protocol(const char *filename,
466 bool allow_protocol_prefix,
467 Error **errp)
468 {
469 BlockDriver *drv1;
470 char protocol[128];
471 int len;
472 const char *p;
473
474 /* TODO Drivers without bdrv_file_open must be specified explicitly */
475
476 /*
477 * XXX(hch): we really should not let host device detection
478 * override an explicit protocol specification, but moving this
479 * later breaks access to device names with colons in them.
480 * Thanks to the brain-dead persistent naming schemes on udev-
481 * based Linux systems those actually are quite common.
482 */
483 drv1 = find_hdev_driver(filename);
484 if (drv1) {
485 return drv1;
486 }
487
488 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
489 return &bdrv_file;
490 }
491
492 p = strchr(filename, ':');
493 assert(p != NULL);
494 len = p - filename;
495 if (len > sizeof(protocol) - 1)
496 len = sizeof(protocol) - 1;
497 memcpy(protocol, filename, len);
498 protocol[len] = '\0';
499 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
500 if (drv1->protocol_name &&
501 !strcmp(drv1->protocol_name, protocol)) {
502 return drv1;
503 }
504 }
505
506 error_setg(errp, "Unknown protocol '%s'", protocol);
507 return NULL;
508 }
509
510 /*
511 * Guess image format by probing its contents.
512 * This is not a good idea when your image is raw (CVE-2008-2004), but
513 * we do it anyway for backward compatibility.
514 *
515 * @buf contains the image's first @buf_size bytes.
516 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
517 * but can be smaller if the image file is smaller)
518 * @filename is its filename.
519 *
520 * For all block drivers, call the bdrv_probe() method to get its
521 * probing score.
522 * Return the first block driver with the highest probing score.
523 */
524 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
525 const char *filename)
526 {
527 int score_max = 0, score;
528 BlockDriver *drv = NULL, *d;
529
530 QLIST_FOREACH(d, &bdrv_drivers, list) {
531 if (d->bdrv_probe) {
532 score = d->bdrv_probe(buf, buf_size, filename);
533 if (score > score_max) {
534 score_max = score;
535 drv = d;
536 }
537 }
538 }
539
540 return drv;
541 }
542
543 static int find_image_format(BlockDriverState *bs, const char *filename,
544 BlockDriver **pdrv, Error **errp)
545 {
546 BlockDriver *drv;
547 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
548 int ret = 0;
549
550 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
551 if (bdrv_is_sg(bs) || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
552 *pdrv = &bdrv_raw;
553 return ret;
554 }
555
556 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
557 if (ret < 0) {
558 error_setg_errno(errp, -ret, "Could not read image for determining its "
559 "format");
560 *pdrv = NULL;
561 return ret;
562 }
563
564 drv = bdrv_probe_all(buf, ret, filename);
565 if (!drv) {
566 error_setg(errp, "Could not determine image format: No compatible "
567 "driver found");
568 ret = -ENOENT;
569 }
570 *pdrv = drv;
571 return ret;
572 }
573
574 /**
575 * Set the current 'total_sectors' value
576 * Return 0 on success, -errno on error.
577 */
578 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
579 {
580 BlockDriver *drv = bs->drv;
581
582 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
583 if (bdrv_is_sg(bs))
584 return 0;
585
586 /* query actual device if possible, otherwise just trust the hint */
587 if (drv->bdrv_getlength) {
588 int64_t length = drv->bdrv_getlength(bs);
589 if (length < 0) {
590 return length;
591 }
592 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
593 }
594
595 bs->total_sectors = hint;
596 return 0;
597 }
598
599 /**
600 * Combines a QDict of new block driver @options with any missing options taken
601 * from @old_options, so that leaving out an option defaults to its old value.
602 */
603 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
604 QDict *old_options)
605 {
606 if (bs->drv && bs->drv->bdrv_join_options) {
607 bs->drv->bdrv_join_options(options, old_options);
608 } else {
609 qdict_join(options, old_options, false);
610 }
611 }
612
613 /**
614 * Set open flags for a given discard mode
615 *
616 * Return 0 on success, -1 if the discard mode was invalid.
617 */
618 int bdrv_parse_discard_flags(const char *mode, int *flags)
619 {
620 *flags &= ~BDRV_O_UNMAP;
621
622 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
623 /* do nothing */
624 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
625 *flags |= BDRV_O_UNMAP;
626 } else {
627 return -1;
628 }
629
630 return 0;
631 }
632
633 /**
634 * Set open flags for a given cache mode
635 *
636 * Return 0 on success, -1 if the cache mode was invalid.
637 */
638 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
639 {
640 *flags &= ~BDRV_O_CACHE_MASK;
641
642 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
643 *writethrough = false;
644 *flags |= BDRV_O_NOCACHE;
645 } else if (!strcmp(mode, "directsync")) {
646 *writethrough = true;
647 *flags |= BDRV_O_NOCACHE;
648 } else if (!strcmp(mode, "writeback")) {
649 *writethrough = false;
650 } else if (!strcmp(mode, "unsafe")) {
651 *writethrough = false;
652 *flags |= BDRV_O_NO_FLUSH;
653 } else if (!strcmp(mode, "writethrough")) {
654 *writethrough = true;
655 } else {
656 return -1;
657 }
658
659 return 0;
660 }
661
662 static void bdrv_child_cb_drained_begin(BdrvChild *child)
663 {
664 BlockDriverState *bs = child->opaque;
665 bdrv_drained_begin(bs);
666 }
667
668 static void bdrv_child_cb_drained_end(BdrvChild *child)
669 {
670 BlockDriverState *bs = child->opaque;
671 bdrv_drained_end(bs);
672 }
673
674 /*
675 * Returns the options and flags that a temporary snapshot should get, based on
676 * the originally requested flags (the originally requested image will have
677 * flags like a backing file)
678 */
679 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
680 int parent_flags, QDict *parent_options)
681 {
682 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
683
684 /* For temporary files, unconditional cache=unsafe is fine */
685 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
686 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
687
688 /* aio=native doesn't work for cache.direct=off, so disable it for the
689 * temporary snapshot */
690 *child_flags &= ~BDRV_O_NATIVE_AIO;
691 }
692
693 /*
694 * Returns the options and flags that bs->file should get if a protocol driver
695 * is expected, based on the given options and flags for the parent BDS
696 */
697 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
698 int parent_flags, QDict *parent_options)
699 {
700 int flags = parent_flags;
701
702 /* Enable protocol handling, disable format probing for bs->file */
703 flags |= BDRV_O_PROTOCOL;
704
705 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
706 * the parent. */
707 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
708 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
709
710 /* Our block drivers take care to send flushes and respect unmap policy,
711 * so we can default to enable both on lower layers regardless of the
712 * corresponding parent options. */
713 flags |= BDRV_O_UNMAP;
714
715 /* Clear flags that only apply to the top layer */
716 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
717 BDRV_O_NO_IO);
718
719 *child_flags = flags;
720 }
721
722 const BdrvChildRole child_file = {
723 .inherit_options = bdrv_inherited_options,
724 .drained_begin = bdrv_child_cb_drained_begin,
725 .drained_end = bdrv_child_cb_drained_end,
726 };
727
728 /*
729 * Returns the options and flags that bs->file should get if the use of formats
730 * (and not only protocols) is permitted for it, based on the given options and
731 * flags for the parent BDS
732 */
733 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
734 int parent_flags, QDict *parent_options)
735 {
736 child_file.inherit_options(child_flags, child_options,
737 parent_flags, parent_options);
738
739 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
740 }
741
742 const BdrvChildRole child_format = {
743 .inherit_options = bdrv_inherited_fmt_options,
744 .drained_begin = bdrv_child_cb_drained_begin,
745 .drained_end = bdrv_child_cb_drained_end,
746 };
747
748 /*
749 * Returns the options and flags that bs->backing should get, based on the
750 * given options and flags for the parent BDS
751 */
752 static void bdrv_backing_options(int *child_flags, QDict *child_options,
753 int parent_flags, QDict *parent_options)
754 {
755 int flags = parent_flags;
756
757 /* The cache mode is inherited unmodified for backing files; except WCE,
758 * which is only applied on the top level (BlockBackend) */
759 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
760 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
761
762 /* backing files always opened read-only */
763 flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
764
765 /* snapshot=on is handled on the top layer */
766 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
767
768 *child_flags = flags;
769 }
770
771 static const BdrvChildRole child_backing = {
772 .inherit_options = bdrv_backing_options,
773 .drained_begin = bdrv_child_cb_drained_begin,
774 .drained_end = bdrv_child_cb_drained_end,
775 };
776
777 static int bdrv_open_flags(BlockDriverState *bs, int flags)
778 {
779 int open_flags = flags;
780
781 /*
782 * Clear flags that are internal to the block layer before opening the
783 * image.
784 */
785 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
786
787 /*
788 * Snapshots should be writable.
789 */
790 if (flags & BDRV_O_TEMPORARY) {
791 open_flags |= BDRV_O_RDWR;
792 }
793
794 return open_flags;
795 }
796
797 static void update_flags_from_options(int *flags, QemuOpts *opts)
798 {
799 *flags &= ~BDRV_O_CACHE_MASK;
800
801 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
802 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
803 *flags |= BDRV_O_NO_FLUSH;
804 }
805
806 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
807 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
808 *flags |= BDRV_O_NOCACHE;
809 }
810 }
811
812 static void update_options_from_flags(QDict *options, int flags)
813 {
814 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
815 qdict_put(options, BDRV_OPT_CACHE_DIRECT,
816 qbool_from_bool(flags & BDRV_O_NOCACHE));
817 }
818 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
819 qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
820 qbool_from_bool(flags & BDRV_O_NO_FLUSH));
821 }
822 }
823
824 static void bdrv_assign_node_name(BlockDriverState *bs,
825 const char *node_name,
826 Error **errp)
827 {
828 char *gen_node_name = NULL;
829
830 if (!node_name) {
831 node_name = gen_node_name = id_generate(ID_BLOCK);
832 } else if (!id_wellformed(node_name)) {
833 /*
834 * Check for empty string or invalid characters, but not if it is
835 * generated (generated names use characters not available to the user)
836 */
837 error_setg(errp, "Invalid node name");
838 return;
839 }
840
841 /* takes care of avoiding namespaces collisions */
842 if (blk_by_name(node_name)) {
843 error_setg(errp, "node-name=%s is conflicting with a device id",
844 node_name);
845 goto out;
846 }
847
848 /* takes care of avoiding duplicates node names */
849 if (bdrv_find_node(node_name)) {
850 error_setg(errp, "Duplicate node name");
851 goto out;
852 }
853
854 /* copy node name into the bs and insert it into the graph list */
855 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
856 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
857 out:
858 g_free(gen_node_name);
859 }
860
861 static QemuOptsList bdrv_runtime_opts = {
862 .name = "bdrv_common",
863 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
864 .desc = {
865 {
866 .name = "node-name",
867 .type = QEMU_OPT_STRING,
868 .help = "Node name of the block device node",
869 },
870 {
871 .name = "driver",
872 .type = QEMU_OPT_STRING,
873 .help = "Block driver to use for the node",
874 },
875 {
876 .name = BDRV_OPT_CACHE_DIRECT,
877 .type = QEMU_OPT_BOOL,
878 .help = "Bypass software writeback cache on the host",
879 },
880 {
881 .name = BDRV_OPT_CACHE_NO_FLUSH,
882 .type = QEMU_OPT_BOOL,
883 .help = "Ignore flush requests",
884 },
885 { /* end of list */ }
886 },
887 };
888
889 /*
890 * Common part for opening disk images and files
891 *
892 * Removes all processed options from *options.
893 */
894 static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
895 QDict *options, Error **errp)
896 {
897 int ret, open_flags;
898 const char *filename;
899 const char *driver_name = NULL;
900 const char *node_name = NULL;
901 QemuOpts *opts;
902 BlockDriver *drv;
903 Error *local_err = NULL;
904
905 assert(bs->file == NULL);
906 assert(options != NULL && bs->options != options);
907
908 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
909 qemu_opts_absorb_qdict(opts, options, &local_err);
910 if (local_err) {
911 error_propagate(errp, local_err);
912 ret = -EINVAL;
913 goto fail_opts;
914 }
915
916 driver_name = qemu_opt_get(opts, "driver");
917 drv = bdrv_find_format(driver_name);
918 assert(drv != NULL);
919
920 if (file != NULL) {
921 filename = file->bs->filename;
922 } else {
923 filename = qdict_get_try_str(options, "filename");
924 }
925
926 if (drv->bdrv_needs_filename && !filename) {
927 error_setg(errp, "The '%s' block driver requires a file name",
928 drv->format_name);
929 ret = -EINVAL;
930 goto fail_opts;
931 }
932
933 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
934 drv->format_name);
935
936 node_name = qemu_opt_get(opts, "node-name");
937 bdrv_assign_node_name(bs, node_name, &local_err);
938 if (local_err) {
939 error_propagate(errp, local_err);
940 ret = -EINVAL;
941 goto fail_opts;
942 }
943
944 bs->request_alignment = drv->bdrv_co_preadv ? 1 : 512;
945 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
946
947 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
948 error_setg(errp,
949 !bs->read_only && bdrv_is_whitelisted(drv, true)
950 ? "Driver '%s' can only be used for read-only devices"
951 : "Driver '%s' is not whitelisted",
952 drv->format_name);
953 ret = -ENOTSUP;
954 goto fail_opts;
955 }
956
957 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
958 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
959 if (!bs->read_only) {
960 bdrv_enable_copy_on_read(bs);
961 } else {
962 error_setg(errp, "Can't use copy-on-read on read-only device");
963 ret = -EINVAL;
964 goto fail_opts;
965 }
966 }
967
968 if (filename != NULL) {
969 pstrcpy(bs->filename, sizeof(bs->filename), filename);
970 } else {
971 bs->filename[0] = '\0';
972 }
973 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
974
975 bs->drv = drv;
976 bs->opaque = g_malloc0(drv->instance_size);
977
978 /* Apply cache mode options */
979 update_flags_from_options(&bs->open_flags, opts);
980
981 /* Open the image, either directly or using a protocol */
982 open_flags = bdrv_open_flags(bs, bs->open_flags);
983 if (drv->bdrv_file_open) {
984 assert(file == NULL);
985 assert(!drv->bdrv_needs_filename || filename != NULL);
986 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
987 } else {
988 if (file == NULL) {
989 error_setg(errp, "Can't use '%s' as a block driver for the "
990 "protocol level", drv->format_name);
991 ret = -EINVAL;
992 goto free_and_fail;
993 }
994 bs->file = file;
995 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
996 }
997
998 if (ret < 0) {
999 if (local_err) {
1000 error_propagate(errp, local_err);
1001 } else if (bs->filename[0]) {
1002 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1003 } else {
1004 error_setg_errno(errp, -ret, "Could not open image");
1005 }
1006 goto free_and_fail;
1007 }
1008
1009 ret = refresh_total_sectors(bs, bs->total_sectors);
1010 if (ret < 0) {
1011 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1012 goto free_and_fail;
1013 }
1014
1015 bdrv_refresh_limits(bs, &local_err);
1016 if (local_err) {
1017 error_propagate(errp, local_err);
1018 ret = -EINVAL;
1019 goto free_and_fail;
1020 }
1021
1022 assert(bdrv_opt_mem_align(bs) != 0);
1023 assert(bdrv_min_mem_align(bs) != 0);
1024 assert(is_power_of_2(bs->request_alignment) || bdrv_is_sg(bs));
1025
1026 qemu_opts_del(opts);
1027 return 0;
1028
1029 free_and_fail:
1030 bs->file = NULL;
1031 g_free(bs->opaque);
1032 bs->opaque = NULL;
1033 bs->drv = NULL;
1034 fail_opts:
1035 qemu_opts_del(opts);
1036 return ret;
1037 }
1038
1039 static QDict *parse_json_filename(const char *filename, Error **errp)
1040 {
1041 QObject *options_obj;
1042 QDict *options;
1043 int ret;
1044
1045 ret = strstart(filename, "json:", &filename);
1046 assert(ret);
1047
1048 options_obj = qobject_from_json(filename);
1049 if (!options_obj) {
1050 error_setg(errp, "Could not parse the JSON options");
1051 return NULL;
1052 }
1053
1054 if (qobject_type(options_obj) != QTYPE_QDICT) {
1055 qobject_decref(options_obj);
1056 error_setg(errp, "Invalid JSON object given");
1057 return NULL;
1058 }
1059
1060 options = qobject_to_qdict(options_obj);
1061 qdict_flatten(options);
1062
1063 return options;
1064 }
1065
1066 static void parse_json_protocol(QDict *options, const char **pfilename,
1067 Error **errp)
1068 {
1069 QDict *json_options;
1070 Error *local_err = NULL;
1071
1072 /* Parse json: pseudo-protocol */
1073 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1074 return;
1075 }
1076
1077 json_options = parse_json_filename(*pfilename, &local_err);
1078 if (local_err) {
1079 error_propagate(errp, local_err);
1080 return;
1081 }
1082
1083 /* Options given in the filename have lower priority than options
1084 * specified directly */
1085 qdict_join(options, json_options, false);
1086 QDECREF(json_options);
1087 *pfilename = NULL;
1088 }
1089
1090 /*
1091 * Fills in default options for opening images and converts the legacy
1092 * filename/flags pair to option QDict entries.
1093 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1094 * block driver has been specified explicitly.
1095 */
1096 static int bdrv_fill_options(QDict **options, const char *filename,
1097 int *flags, Error **errp)
1098 {
1099 const char *drvname;
1100 bool protocol = *flags & BDRV_O_PROTOCOL;
1101 bool parse_filename = false;
1102 BlockDriver *drv = NULL;
1103 Error *local_err = NULL;
1104
1105 drvname = qdict_get_try_str(*options, "driver");
1106 if (drvname) {
1107 drv = bdrv_find_format(drvname);
1108 if (!drv) {
1109 error_setg(errp, "Unknown driver '%s'", drvname);
1110 return -ENOENT;
1111 }
1112 /* If the user has explicitly specified the driver, this choice should
1113 * override the BDRV_O_PROTOCOL flag */
1114 protocol = drv->bdrv_file_open;
1115 }
1116
1117 if (protocol) {
1118 *flags |= BDRV_O_PROTOCOL;
1119 } else {
1120 *flags &= ~BDRV_O_PROTOCOL;
1121 }
1122
1123 /* Translate cache options from flags into options */
1124 update_options_from_flags(*options, *flags);
1125
1126 /* Fetch the file name from the options QDict if necessary */
1127 if (protocol && filename) {
1128 if (!qdict_haskey(*options, "filename")) {
1129 qdict_put(*options, "filename", qstring_from_str(filename));
1130 parse_filename = true;
1131 } else {
1132 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1133 "the same time");
1134 return -EINVAL;
1135 }
1136 }
1137
1138 /* Find the right block driver */
1139 filename = qdict_get_try_str(*options, "filename");
1140
1141 if (!drvname && protocol) {
1142 if (filename) {
1143 drv = bdrv_find_protocol(filename, parse_filename, errp);
1144 if (!drv) {
1145 return -EINVAL;
1146 }
1147
1148 drvname = drv->format_name;
1149 qdict_put(*options, "driver", qstring_from_str(drvname));
1150 } else {
1151 error_setg(errp, "Must specify either driver or file");
1152 return -EINVAL;
1153 }
1154 }
1155
1156 assert(drv || !protocol);
1157
1158 /* Driver-specific filename parsing */
1159 if (drv && drv->bdrv_parse_filename && parse_filename) {
1160 drv->bdrv_parse_filename(filename, *options, &local_err);
1161 if (local_err) {
1162 error_propagate(errp, local_err);
1163 return -EINVAL;
1164 }
1165
1166 if (!drv->bdrv_needs_filename) {
1167 qdict_del(*options, "filename");
1168 }
1169 }
1170
1171 return 0;
1172 }
1173
1174 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1175 {
1176 BlockDriverState *old_bs = child->bs;
1177
1178 if (old_bs) {
1179 if (old_bs->quiesce_counter && child->role->drained_end) {
1180 child->role->drained_end(child);
1181 }
1182 QLIST_REMOVE(child, next_parent);
1183 }
1184
1185 child->bs = new_bs;
1186
1187 if (new_bs) {
1188 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1189 if (new_bs->quiesce_counter && child->role->drained_begin) {
1190 child->role->drained_begin(child);
1191 }
1192 }
1193 }
1194
1195 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1196 const char *child_name,
1197 const BdrvChildRole *child_role,
1198 void *opaque)
1199 {
1200 BdrvChild *child = g_new(BdrvChild, 1);
1201 *child = (BdrvChild) {
1202 .bs = NULL,
1203 .name = g_strdup(child_name),
1204 .role = child_role,
1205 .opaque = opaque,
1206 };
1207
1208 bdrv_replace_child(child, child_bs);
1209
1210 return child;
1211 }
1212
1213 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1214 BlockDriverState *child_bs,
1215 const char *child_name,
1216 const BdrvChildRole *child_role)
1217 {
1218 BdrvChild *child = bdrv_root_attach_child(child_bs, child_name, child_role,
1219 parent_bs);
1220 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1221 return child;
1222 }
1223
1224 static void bdrv_detach_child(BdrvChild *child)
1225 {
1226 if (child->next.le_prev) {
1227 QLIST_REMOVE(child, next);
1228 child->next.le_prev = NULL;
1229 }
1230
1231 bdrv_replace_child(child, NULL);
1232
1233 g_free(child->name);
1234 g_free(child);
1235 }
1236
1237 void bdrv_root_unref_child(BdrvChild *child)
1238 {
1239 BlockDriverState *child_bs;
1240
1241 child_bs = child->bs;
1242 bdrv_detach_child(child);
1243 bdrv_unref(child_bs);
1244 }
1245
1246 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1247 {
1248 if (child == NULL) {
1249 return;
1250 }
1251
1252 if (child->bs->inherits_from == parent) {
1253 child->bs->inherits_from = NULL;
1254 }
1255
1256 bdrv_root_unref_child(child);
1257 }
1258
1259
1260 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1261 {
1262 BdrvChild *c;
1263 QLIST_FOREACH(c, &bs->parents, next_parent) {
1264 if (c->role->change_media) {
1265 c->role->change_media(c, load);
1266 }
1267 }
1268 }
1269
1270 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1271 {
1272 BdrvChild *c;
1273 QLIST_FOREACH(c, &bs->parents, next_parent) {
1274 if (c->role->resize) {
1275 c->role->resize(c);
1276 }
1277 }
1278 }
1279
1280 /*
1281 * Sets the backing file link of a BDS. A new reference is created; callers
1282 * which don't need their own reference any more must call bdrv_unref().
1283 */
1284 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1285 {
1286 if (backing_hd) {
1287 bdrv_ref(backing_hd);
1288 }
1289
1290 if (bs->backing) {
1291 assert(bs->backing_blocker);
1292 bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker);
1293 bdrv_unref_child(bs, bs->backing);
1294 } else if (backing_hd) {
1295 error_setg(&bs->backing_blocker,
1296 "node is used as backing hd of '%s'",
1297 bdrv_get_device_or_node_name(bs));
1298 }
1299
1300 if (!backing_hd) {
1301 error_free(bs->backing_blocker);
1302 bs->backing_blocker = NULL;
1303 bs->backing = NULL;
1304 goto out;
1305 }
1306 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing);
1307 bs->open_flags &= ~BDRV_O_NO_BACKING;
1308 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1309 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1310 backing_hd->drv ? backing_hd->drv->format_name : "");
1311
1312 bdrv_op_block_all(backing_hd, bs->backing_blocker);
1313 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1314 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1315 bs->backing_blocker);
1316 out:
1317 bdrv_refresh_limits(bs, NULL);
1318 }
1319
1320 /*
1321 * Opens the backing file for a BlockDriverState if not yet open
1322 *
1323 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1324 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1325 * itself, all options starting with "${bdref_key}." are considered part of the
1326 * BlockdevRef.
1327 *
1328 * TODO Can this be unified with bdrv_open_image()?
1329 */
1330 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1331 const char *bdref_key, Error **errp)
1332 {
1333 char *backing_filename = g_malloc0(PATH_MAX);
1334 char *bdref_key_dot;
1335 const char *reference = NULL;
1336 int ret = 0;
1337 BlockDriverState *backing_hd;
1338 QDict *options;
1339 QDict *tmp_parent_options = NULL;
1340 Error *local_err = NULL;
1341
1342 if (bs->backing != NULL) {
1343 goto free_exit;
1344 }
1345
1346 /* NULL means an empty set of options */
1347 if (parent_options == NULL) {
1348 tmp_parent_options = qdict_new();
1349 parent_options = tmp_parent_options;
1350 }
1351
1352 bs->open_flags &= ~BDRV_O_NO_BACKING;
1353
1354 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1355 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
1356 g_free(bdref_key_dot);
1357
1358 reference = qdict_get_try_str(parent_options, bdref_key);
1359 if (reference || qdict_haskey(options, "file.filename")) {
1360 backing_filename[0] = '\0';
1361 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1362 QDECREF(options);
1363 goto free_exit;
1364 } else {
1365 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
1366 &local_err);
1367 if (local_err) {
1368 ret = -EINVAL;
1369 error_propagate(errp, local_err);
1370 QDECREF(options);
1371 goto free_exit;
1372 }
1373 }
1374
1375 if (!bs->drv || !bs->drv->supports_backing) {
1376 ret = -EINVAL;
1377 error_setg(errp, "Driver doesn't support backing files");
1378 QDECREF(options);
1379 goto free_exit;
1380 }
1381
1382 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
1383 qdict_put(options, "driver", qstring_from_str(bs->backing_format));
1384 }
1385
1386 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
1387 reference, options, 0, bs, &child_backing,
1388 errp);
1389 if (!backing_hd) {
1390 bs->open_flags |= BDRV_O_NO_BACKING;
1391 error_prepend(errp, "Could not open backing file: ");
1392 ret = -EINVAL;
1393 goto free_exit;
1394 }
1395
1396 /* Hook up the backing file link; drop our reference, bs owns the
1397 * backing_hd reference now */
1398 bdrv_set_backing_hd(bs, backing_hd);
1399 bdrv_unref(backing_hd);
1400
1401 qdict_del(parent_options, bdref_key);
1402
1403 free_exit:
1404 g_free(backing_filename);
1405 QDECREF(tmp_parent_options);
1406 return ret;
1407 }
1408
1409 /*
1410 * Opens a disk image whose options are given as BlockdevRef in another block
1411 * device's options.
1412 *
1413 * If allow_none is true, no image will be opened if filename is false and no
1414 * BlockdevRef is given. NULL will be returned, but errp remains unset.
1415 *
1416 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1417 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1418 * itself, all options starting with "${bdref_key}." are considered part of the
1419 * BlockdevRef.
1420 *
1421 * The BlockdevRef will be removed from the options QDict.
1422 */
1423 BdrvChild *bdrv_open_child(const char *filename,
1424 QDict *options, const char *bdref_key,
1425 BlockDriverState* parent,
1426 const BdrvChildRole *child_role,
1427 bool allow_none, Error **errp)
1428 {
1429 BdrvChild *c = NULL;
1430 BlockDriverState *bs;
1431 QDict *image_options;
1432 char *bdref_key_dot;
1433 const char *reference;
1434
1435 assert(child_role != NULL);
1436
1437 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1438 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1439 g_free(bdref_key_dot);
1440
1441 reference = qdict_get_try_str(options, bdref_key);
1442 if (!filename && !reference && !qdict_size(image_options)) {
1443 if (!allow_none) {
1444 error_setg(errp, "A block device must be specified for \"%s\"",
1445 bdref_key);
1446 }
1447 QDECREF(image_options);
1448 goto done;
1449 }
1450
1451 bs = bdrv_open_inherit(filename, reference, image_options, 0,
1452 parent, child_role, errp);
1453 if (!bs) {
1454 goto done;
1455 }
1456
1457 c = bdrv_attach_child(parent, bs, bdref_key, child_role);
1458
1459 done:
1460 qdict_del(options, bdref_key);
1461 return c;
1462 }
1463
1464 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
1465 int flags,
1466 QDict *snapshot_options,
1467 Error **errp)
1468 {
1469 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1470 char *tmp_filename = g_malloc0(PATH_MAX + 1);
1471 int64_t total_size;
1472 QemuOpts *opts = NULL;
1473 BlockDriverState *bs_snapshot;
1474 int ret;
1475
1476 /* if snapshot, we create a temporary backing file and open it
1477 instead of opening 'filename' directly */
1478
1479 /* Get the required size from the image */
1480 total_size = bdrv_getlength(bs);
1481 if (total_size < 0) {
1482 error_setg_errno(errp, -total_size, "Could not get image size");
1483 goto out;
1484 }
1485
1486 /* Create the temporary image */
1487 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1488 if (ret < 0) {
1489 error_setg_errno(errp, -ret, "Could not get temporary filename");
1490 goto out;
1491 }
1492
1493 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
1494 &error_abort);
1495 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
1496 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
1497 qemu_opts_del(opts);
1498 if (ret < 0) {
1499 error_prepend(errp, "Could not create temporary overlay '%s': ",
1500 tmp_filename);
1501 goto out;
1502 }
1503
1504 /* Prepare options QDict for the temporary file */
1505 qdict_put(snapshot_options, "file.driver",
1506 qstring_from_str("file"));
1507 qdict_put(snapshot_options, "file.filename",
1508 qstring_from_str(tmp_filename));
1509 qdict_put(snapshot_options, "driver",
1510 qstring_from_str("qcow2"));
1511
1512 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
1513 snapshot_options = NULL;
1514 if (!bs_snapshot) {
1515 ret = -EINVAL;
1516 goto out;
1517 }
1518
1519 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
1520 * call bdrv_unref() on it), so in order to be able to return one, we have
1521 * to increase bs_snapshot's refcount here */
1522 bdrv_ref(bs_snapshot);
1523 bdrv_append(bs_snapshot, bs);
1524
1525 g_free(tmp_filename);
1526 return bs_snapshot;
1527
1528 out:
1529 QDECREF(snapshot_options);
1530 g_free(tmp_filename);
1531 return NULL;
1532 }
1533
1534 /*
1535 * Opens a disk image (raw, qcow2, vmdk, ...)
1536 *
1537 * options is a QDict of options to pass to the block drivers, or NULL for an
1538 * empty set of options. The reference to the QDict belongs to the block layer
1539 * after the call (even on failure), so if the caller intends to reuse the
1540 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1541 *
1542 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1543 * If it is not NULL, the referenced BDS will be reused.
1544 *
1545 * The reference parameter may be used to specify an existing block device which
1546 * should be opened. If specified, neither options nor a filename may be given,
1547 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1548 */
1549 static BlockDriverState *bdrv_open_inherit(const char *filename,
1550 const char *reference,
1551 QDict *options, int flags,
1552 BlockDriverState *parent,
1553 const BdrvChildRole *child_role,
1554 Error **errp)
1555 {
1556 int ret;
1557 BdrvChild *file = NULL;
1558 BlockDriverState *bs;
1559 BlockDriver *drv = NULL;
1560 const char *drvname;
1561 const char *backing;
1562 Error *local_err = NULL;
1563 QDict *snapshot_options = NULL;
1564 int snapshot_flags = 0;
1565
1566 assert(!child_role || !flags);
1567 assert(!child_role == !parent);
1568
1569 if (reference) {
1570 bool options_non_empty = options ? qdict_size(options) : false;
1571 QDECREF(options);
1572
1573 if (filename || options_non_empty) {
1574 error_setg(errp, "Cannot reference an existing block device with "
1575 "additional options or a new filename");
1576 return NULL;
1577 }
1578
1579 bs = bdrv_lookup_bs(reference, reference, errp);
1580 if (!bs) {
1581 return NULL;
1582 }
1583
1584 bdrv_ref(bs);
1585 return bs;
1586 }
1587
1588 bs = bdrv_new();
1589
1590 /* NULL means an empty set of options */
1591 if (options == NULL) {
1592 options = qdict_new();
1593 }
1594
1595 /* json: syntax counts as explicit options, as if in the QDict */
1596 parse_json_protocol(options, &filename, &local_err);
1597 if (local_err) {
1598 goto fail;
1599 }
1600
1601 bs->explicit_options = qdict_clone_shallow(options);
1602
1603 if (child_role) {
1604 bs->inherits_from = parent;
1605 child_role->inherit_options(&flags, options,
1606 parent->open_flags, parent->options);
1607 }
1608
1609 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
1610 if (local_err) {
1611 goto fail;
1612 }
1613
1614 bs->open_flags = flags;
1615 bs->options = options;
1616 options = qdict_clone_shallow(options);
1617
1618 /* Find the right image format driver */
1619 drvname = qdict_get_try_str(options, "driver");
1620 if (drvname) {
1621 drv = bdrv_find_format(drvname);
1622 if (!drv) {
1623 error_setg(errp, "Unknown driver: '%s'", drvname);
1624 goto fail;
1625 }
1626 }
1627
1628 assert(drvname || !(flags & BDRV_O_PROTOCOL));
1629
1630 backing = qdict_get_try_str(options, "backing");
1631 if (backing && *backing == '\0') {
1632 flags |= BDRV_O_NO_BACKING;
1633 qdict_del(options, "backing");
1634 }
1635
1636 /* Open image file without format layer */
1637 if ((flags & BDRV_O_PROTOCOL) == 0) {
1638 if (flags & BDRV_O_RDWR) {
1639 flags |= BDRV_O_ALLOW_RDWR;
1640 }
1641 if (flags & BDRV_O_SNAPSHOT) {
1642 snapshot_options = qdict_new();
1643 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
1644 flags, options);
1645 bdrv_backing_options(&flags, options, flags, options);
1646 }
1647
1648 bs->open_flags = flags;
1649
1650 file = bdrv_open_child(filename, options, "file", bs,
1651 &child_file, true, &local_err);
1652 if (local_err) {
1653 goto fail;
1654 }
1655 }
1656
1657 /* Image format probing */
1658 bs->probed = !drv;
1659 if (!drv && file) {
1660 ret = find_image_format(file->bs, filename, &drv, &local_err);
1661 if (ret < 0) {
1662 goto fail;
1663 }
1664 /*
1665 * This option update would logically belong in bdrv_fill_options(),
1666 * but we first need to open bs->file for the probing to work, while
1667 * opening bs->file already requires the (mostly) final set of options
1668 * so that cache mode etc. can be inherited.
1669 *
1670 * Adding the driver later is somewhat ugly, but it's not an option
1671 * that would ever be inherited, so it's correct. We just need to make
1672 * sure to update both bs->options (which has the full effective
1673 * options for bs) and options (which has file.* already removed).
1674 */
1675 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
1676 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1677 } else if (!drv) {
1678 error_setg(errp, "Must specify either driver or file");
1679 goto fail;
1680 }
1681
1682 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
1683 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
1684 /* file must be NULL if a protocol BDS is about to be created
1685 * (the inverse results in an error message from bdrv_open_common()) */
1686 assert(!(flags & BDRV_O_PROTOCOL) || !file);
1687
1688 /* Open the image */
1689 ret = bdrv_open_common(bs, file, options, &local_err);
1690 if (ret < 0) {
1691 goto fail;
1692 }
1693
1694 if (file && (bs->file != file)) {
1695 bdrv_unref_child(bs, file);
1696 file = NULL;
1697 }
1698
1699 /* If there is a backing file, use it */
1700 if ((flags & BDRV_O_NO_BACKING) == 0) {
1701 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
1702 if (ret < 0) {
1703 goto close_and_fail;
1704 }
1705 }
1706
1707 bdrv_refresh_filename(bs);
1708
1709 /* Check if any unknown options were used */
1710 if (options && (qdict_size(options) != 0)) {
1711 const QDictEntry *entry = qdict_first(options);
1712 if (flags & BDRV_O_PROTOCOL) {
1713 error_setg(errp, "Block protocol '%s' doesn't support the option "
1714 "'%s'", drv->format_name, entry->key);
1715 } else {
1716 error_setg(errp,
1717 "Block format '%s' does not support the option '%s'",
1718 drv->format_name, entry->key);
1719 }
1720
1721 goto close_and_fail;
1722 }
1723
1724 if (!bdrv_key_required(bs)) {
1725 bdrv_parent_cb_change_media(bs, true);
1726 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1727 && !runstate_check(RUN_STATE_INMIGRATE)
1728 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1729 error_setg(errp,
1730 "Guest must be stopped for opening of encrypted image");
1731 goto close_and_fail;
1732 }
1733
1734 QDECREF(options);
1735
1736 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1737 * temporary snapshot afterwards. */
1738 if (snapshot_flags) {
1739 BlockDriverState *snapshot_bs;
1740 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
1741 snapshot_options, &local_err);
1742 snapshot_options = NULL;
1743 if (local_err) {
1744 goto close_and_fail;
1745 }
1746 /* We are not going to return bs but the overlay on top of it
1747 * (snapshot_bs); thus, we have to drop the strong reference to bs
1748 * (which we obtained by calling bdrv_new()). bs will not be deleted,
1749 * though, because the overlay still has a reference to it. */
1750 bdrv_unref(bs);
1751 bs = snapshot_bs;
1752 }
1753
1754 return bs;
1755
1756 fail:
1757 if (file != NULL) {
1758 bdrv_unref_child(bs, file);
1759 }
1760 QDECREF(snapshot_options);
1761 QDECREF(bs->explicit_options);
1762 QDECREF(bs->options);
1763 QDECREF(options);
1764 bs->options = NULL;
1765 bdrv_unref(bs);
1766 if (local_err) {
1767 error_propagate(errp, local_err);
1768 }
1769 return NULL;
1770
1771 close_and_fail:
1772 bdrv_unref(bs);
1773 QDECREF(snapshot_options);
1774 QDECREF(options);
1775 if (local_err) {
1776 error_propagate(errp, local_err);
1777 }
1778 return NULL;
1779 }
1780
1781 BlockDriverState *bdrv_open(const char *filename, const char *reference,
1782 QDict *options, int flags, Error **errp)
1783 {
1784 return bdrv_open_inherit(filename, reference, options, flags, NULL,
1785 NULL, errp);
1786 }
1787
1788 typedef struct BlockReopenQueueEntry {
1789 bool prepared;
1790 BDRVReopenState state;
1791 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1792 } BlockReopenQueueEntry;
1793
1794 /*
1795 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1796 * reopen of multiple devices.
1797 *
1798 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1799 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1800 * be created and initialized. This newly created BlockReopenQueue should be
1801 * passed back in for subsequent calls that are intended to be of the same
1802 * atomic 'set'.
1803 *
1804 * bs is the BlockDriverState to add to the reopen queue.
1805 *
1806 * options contains the changed options for the associated bs
1807 * (the BlockReopenQueue takes ownership)
1808 *
1809 * flags contains the open flags for the associated bs
1810 *
1811 * returns a pointer to bs_queue, which is either the newly allocated
1812 * bs_queue, or the existing bs_queue being used.
1813 *
1814 */
1815 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
1816 BlockDriverState *bs,
1817 QDict *options,
1818 int flags,
1819 const BdrvChildRole *role,
1820 QDict *parent_options,
1821 int parent_flags)
1822 {
1823 assert(bs != NULL);
1824
1825 BlockReopenQueueEntry *bs_entry;
1826 BdrvChild *child;
1827 QDict *old_options, *explicit_options;
1828
1829 if (bs_queue == NULL) {
1830 bs_queue = g_new0(BlockReopenQueue, 1);
1831 QSIMPLEQ_INIT(bs_queue);
1832 }
1833
1834 if (!options) {
1835 options = qdict_new();
1836 }
1837
1838 /*
1839 * Precedence of options:
1840 * 1. Explicitly passed in options (highest)
1841 * 2. Set in flags (only for top level)
1842 * 3. Retained from explicitly set options of bs
1843 * 4. Inherited from parent node
1844 * 5. Retained from effective options of bs
1845 */
1846
1847 if (!parent_options) {
1848 /*
1849 * Any setting represented by flags is always updated. If the
1850 * corresponding QDict option is set, it takes precedence. Otherwise
1851 * the flag is translated into a QDict option. The old setting of bs is
1852 * not considered.
1853 */
1854 update_options_from_flags(options, flags);
1855 }
1856
1857 /* Old explicitly set values (don't overwrite by inherited value) */
1858 old_options = qdict_clone_shallow(bs->explicit_options);
1859 bdrv_join_options(bs, options, old_options);
1860 QDECREF(old_options);
1861
1862 explicit_options = qdict_clone_shallow(options);
1863
1864 /* Inherit from parent node */
1865 if (parent_options) {
1866 assert(!flags);
1867 role->inherit_options(&flags, options, parent_flags, parent_options);
1868 }
1869
1870 /* Old values are used for options that aren't set yet */
1871 old_options = qdict_clone_shallow(bs->options);
1872 bdrv_join_options(bs, options, old_options);
1873 QDECREF(old_options);
1874
1875 /* bdrv_open() masks this flag out */
1876 flags &= ~BDRV_O_PROTOCOL;
1877
1878 QLIST_FOREACH(child, &bs->children, next) {
1879 QDict *new_child_options;
1880 char *child_key_dot;
1881
1882 /* reopen can only change the options of block devices that were
1883 * implicitly created and inherited options. For other (referenced)
1884 * block devices, a syntax like "backing.foo" results in an error. */
1885 if (child->bs->inherits_from != bs) {
1886 continue;
1887 }
1888
1889 child_key_dot = g_strdup_printf("%s.", child->name);
1890 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
1891 g_free(child_key_dot);
1892
1893 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
1894 child->role, options, flags);
1895 }
1896
1897 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1898 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1899
1900 bs_entry->state.bs = bs;
1901 bs_entry->state.options = options;
1902 bs_entry->state.explicit_options = explicit_options;
1903 bs_entry->state.flags = flags;
1904
1905 return bs_queue;
1906 }
1907
1908 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1909 BlockDriverState *bs,
1910 QDict *options, int flags)
1911 {
1912 return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
1913 NULL, NULL, 0);
1914 }
1915
1916 /*
1917 * Reopen multiple BlockDriverStates atomically & transactionally.
1918 *
1919 * The queue passed in (bs_queue) must have been built up previous
1920 * via bdrv_reopen_queue().
1921 *
1922 * Reopens all BDS specified in the queue, with the appropriate
1923 * flags. All devices are prepared for reopen, and failure of any
1924 * device will cause all device changes to be abandonded, and intermediate
1925 * data cleaned up.
1926 *
1927 * If all devices prepare successfully, then the changes are committed
1928 * to all devices.
1929 *
1930 */
1931 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1932 {
1933 int ret = -1;
1934 BlockReopenQueueEntry *bs_entry, *next;
1935 Error *local_err = NULL;
1936
1937 assert(bs_queue != NULL);
1938
1939 bdrv_drain_all();
1940
1941 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1942 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1943 error_propagate(errp, local_err);
1944 goto cleanup;
1945 }
1946 bs_entry->prepared = true;
1947 }
1948
1949 /* If we reach this point, we have success and just need to apply the
1950 * changes
1951 */
1952 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1953 bdrv_reopen_commit(&bs_entry->state);
1954 }
1955
1956 ret = 0;
1957
1958 cleanup:
1959 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1960 if (ret && bs_entry->prepared) {
1961 bdrv_reopen_abort(&bs_entry->state);
1962 } else if (ret) {
1963 QDECREF(bs_entry->state.explicit_options);
1964 }
1965 QDECREF(bs_entry->state.options);
1966 g_free(bs_entry);
1967 }
1968 g_free(bs_queue);
1969 return ret;
1970 }
1971
1972
1973 /* Reopen a single BlockDriverState with the specified flags. */
1974 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1975 {
1976 int ret = -1;
1977 Error *local_err = NULL;
1978 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
1979
1980 ret = bdrv_reopen_multiple(queue, &local_err);
1981 if (local_err != NULL) {
1982 error_propagate(errp, local_err);
1983 }
1984 return ret;
1985 }
1986
1987
1988 /*
1989 * Prepares a BlockDriverState for reopen. All changes are staged in the
1990 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1991 * the block driver layer .bdrv_reopen_prepare()
1992 *
1993 * bs is the BlockDriverState to reopen
1994 * flags are the new open flags
1995 * queue is the reopen queue
1996 *
1997 * Returns 0 on success, non-zero on error. On error errp will be set
1998 * as well.
1999 *
2000 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2001 * It is the responsibility of the caller to then call the abort() or
2002 * commit() for any other BDS that have been left in a prepare() state
2003 *
2004 */
2005 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2006 Error **errp)
2007 {
2008 int ret = -1;
2009 Error *local_err = NULL;
2010 BlockDriver *drv;
2011 QemuOpts *opts;
2012 const char *value;
2013
2014 assert(reopen_state != NULL);
2015 assert(reopen_state->bs->drv != NULL);
2016 drv = reopen_state->bs->drv;
2017
2018 /* Process generic block layer options */
2019 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2020 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2021 if (local_err) {
2022 error_propagate(errp, local_err);
2023 ret = -EINVAL;
2024 goto error;
2025 }
2026
2027 update_flags_from_options(&reopen_state->flags, opts);
2028
2029 /* node-name and driver must be unchanged. Put them back into the QDict, so
2030 * that they are checked at the end of this function. */
2031 value = qemu_opt_get(opts, "node-name");
2032 if (value) {
2033 qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2034 }
2035
2036 value = qemu_opt_get(opts, "driver");
2037 if (value) {
2038 qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2039 }
2040
2041 /* if we are to stay read-only, do not allow permission change
2042 * to r/w */
2043 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2044 reopen_state->flags & BDRV_O_RDWR) {
2045 error_setg(errp, "Node '%s' is read only",
2046 bdrv_get_device_or_node_name(reopen_state->bs));
2047 goto error;
2048 }
2049
2050
2051 ret = bdrv_flush(reopen_state->bs);
2052 if (ret) {
2053 error_setg_errno(errp, -ret, "Error flushing drive");
2054 goto error;
2055 }
2056
2057 if (drv->bdrv_reopen_prepare) {
2058 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2059 if (ret) {
2060 if (local_err != NULL) {
2061 error_propagate(errp, local_err);
2062 } else {
2063 error_setg(errp, "failed while preparing to reopen image '%s'",
2064 reopen_state->bs->filename);
2065 }
2066 goto error;
2067 }
2068 } else {
2069 /* It is currently mandatory to have a bdrv_reopen_prepare()
2070 * handler for each supported drv. */
2071 error_setg(errp, "Block format '%s' used by node '%s' "
2072 "does not support reopening files", drv->format_name,
2073 bdrv_get_device_or_node_name(reopen_state->bs));
2074 ret = -1;
2075 goto error;
2076 }
2077
2078 /* Options that are not handled are only okay if they are unchanged
2079 * compared to the old state. It is expected that some options are only
2080 * used for the initial open, but not reopen (e.g. filename) */
2081 if (qdict_size(reopen_state->options)) {
2082 const QDictEntry *entry = qdict_first(reopen_state->options);
2083
2084 do {
2085 QString *new_obj = qobject_to_qstring(entry->value);
2086 const char *new = qstring_get_str(new_obj);
2087 const char *old = qdict_get_try_str(reopen_state->bs->options,
2088 entry->key);
2089
2090 if (!old || strcmp(new, old)) {
2091 error_setg(errp, "Cannot change the option '%s'", entry->key);
2092 ret = -EINVAL;
2093 goto error;
2094 }
2095 } while ((entry = qdict_next(reopen_state->options, entry)));
2096 }
2097
2098 ret = 0;
2099
2100 error:
2101 qemu_opts_del(opts);
2102 return ret;
2103 }
2104
2105 /*
2106 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2107 * makes them final by swapping the staging BlockDriverState contents into
2108 * the active BlockDriverState contents.
2109 */
2110 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2111 {
2112 BlockDriver *drv;
2113
2114 assert(reopen_state != NULL);
2115 drv = reopen_state->bs->drv;
2116 assert(drv != NULL);
2117
2118 /* If there are any driver level actions to take */
2119 if (drv->bdrv_reopen_commit) {
2120 drv->bdrv_reopen_commit(reopen_state);
2121 }
2122
2123 /* set BDS specific flags now */
2124 QDECREF(reopen_state->bs->explicit_options);
2125
2126 reopen_state->bs->explicit_options = reopen_state->explicit_options;
2127 reopen_state->bs->open_flags = reopen_state->flags;
2128 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2129
2130 bdrv_refresh_limits(reopen_state->bs, NULL);
2131 }
2132
2133 /*
2134 * Abort the reopen, and delete and free the staged changes in
2135 * reopen_state
2136 */
2137 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2138 {
2139 BlockDriver *drv;
2140
2141 assert(reopen_state != NULL);
2142 drv = reopen_state->bs->drv;
2143 assert(drv != NULL);
2144
2145 if (drv->bdrv_reopen_abort) {
2146 drv->bdrv_reopen_abort(reopen_state);
2147 }
2148
2149 QDECREF(reopen_state->explicit_options);
2150 }
2151
2152
2153 static void bdrv_close(BlockDriverState *bs)
2154 {
2155 BdrvAioNotifier *ban, *ban_next;
2156
2157 assert(!bs->job);
2158 assert(!bs->refcnt);
2159
2160 bdrv_drained_begin(bs); /* complete I/O */
2161 bdrv_flush(bs);
2162 bdrv_drain(bs); /* in case flush left pending I/O */
2163
2164 bdrv_release_named_dirty_bitmaps(bs);
2165 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2166
2167 if (bs->drv) {
2168 BdrvChild *child, *next;
2169
2170 bs->drv->bdrv_close(bs);
2171 bs->drv = NULL;
2172
2173 bdrv_set_backing_hd(bs, NULL);
2174
2175 if (bs->file != NULL) {
2176 bdrv_unref_child(bs, bs->file);
2177 bs->file = NULL;
2178 }
2179
2180 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2181 /* TODO Remove bdrv_unref() from drivers' close function and use
2182 * bdrv_unref_child() here */
2183 if (child->bs->inherits_from == bs) {
2184 child->bs->inherits_from = NULL;
2185 }
2186 bdrv_detach_child(child);
2187 }
2188
2189 g_free(bs->opaque);
2190 bs->opaque = NULL;
2191 bs->copy_on_read = 0;
2192 bs->backing_file[0] = '\0';
2193 bs->backing_format[0] = '\0';
2194 bs->total_sectors = 0;
2195 bs->encrypted = 0;
2196 bs->valid_key = 0;
2197 bs->sg = 0;
2198 QDECREF(bs->options);
2199 QDECREF(bs->explicit_options);
2200 bs->options = NULL;
2201 QDECREF(bs->full_open_options);
2202 bs->full_open_options = NULL;
2203 }
2204
2205 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2206 g_free(ban);
2207 }
2208 QLIST_INIT(&bs->aio_notifiers);
2209 bdrv_drained_end(bs);
2210 }
2211
2212 void bdrv_close_all(void)
2213 {
2214 block_job_cancel_sync_all();
2215
2216 /* Drop references from requests still in flight, such as canceled block
2217 * jobs whose AIO context has not been polled yet */
2218 bdrv_drain_all();
2219
2220 blk_remove_all_bs();
2221 blockdev_close_all_bdrv_states();
2222
2223 assert(QTAILQ_EMPTY(&all_bdrv_states));
2224 }
2225
2226 static void change_parent_backing_link(BlockDriverState *from,
2227 BlockDriverState *to)
2228 {
2229 BdrvChild *c, *next, *to_c;
2230
2231 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
2232 if (c->role == &child_backing) {
2233 /* @from is generally not allowed to be a backing file, except for
2234 * when @to is the overlay. In that case, @from may not be replaced
2235 * by @to as @to's backing node. */
2236 QLIST_FOREACH(to_c, &to->children, next) {
2237 if (to_c == c) {
2238 break;
2239 }
2240 }
2241 if (to_c) {
2242 continue;
2243 }
2244 }
2245
2246 assert(c->role != &child_backing);
2247 bdrv_ref(to);
2248 bdrv_replace_child(c, to);
2249 bdrv_unref(from);
2250 }
2251 }
2252
2253 /*
2254 * Add new bs contents at the top of an image chain while the chain is
2255 * live, while keeping required fields on the top layer.
2256 *
2257 * This will modify the BlockDriverState fields, and swap contents
2258 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2259 *
2260 * bs_new must not be attached to a BlockBackend.
2261 *
2262 * This function does not create any image files.
2263 *
2264 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
2265 * that's what the callers commonly need. bs_new will be referenced by the old
2266 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
2267 * reference of its own, it must call bdrv_ref().
2268 */
2269 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2270 {
2271 assert(!bdrv_requests_pending(bs_top));
2272 assert(!bdrv_requests_pending(bs_new));
2273
2274 bdrv_ref(bs_top);
2275
2276 change_parent_backing_link(bs_top, bs_new);
2277 bdrv_set_backing_hd(bs_new, bs_top);
2278 bdrv_unref(bs_top);
2279
2280 /* bs_new is now referenced by its new parents, we don't need the
2281 * additional reference any more. */
2282 bdrv_unref(bs_new);
2283 }
2284
2285 void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
2286 {
2287 assert(!bdrv_requests_pending(old));
2288 assert(!bdrv_requests_pending(new));
2289
2290 bdrv_ref(old);
2291
2292 change_parent_backing_link(old, new);
2293
2294 bdrv_unref(old);
2295 }
2296
2297 static void bdrv_delete(BlockDriverState *bs)
2298 {
2299 assert(!bs->job);
2300 assert(bdrv_op_blocker_is_empty(bs));
2301 assert(!bs->refcnt);
2302
2303 bdrv_close(bs);
2304
2305 /* remove from list, if necessary */
2306 if (bs->node_name[0] != '\0') {
2307 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
2308 }
2309 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
2310
2311 g_free(bs);
2312 }
2313
2314 /*
2315 * Run consistency checks on an image
2316 *
2317 * Returns 0 if the check could be completed (it doesn't mean that the image is
2318 * free of errors) or -errno when an internal error occurred. The results of the
2319 * check are stored in res.
2320 */
2321 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2322 {
2323 if (bs->drv == NULL) {
2324 return -ENOMEDIUM;
2325 }
2326 if (bs->drv->bdrv_check == NULL) {
2327 return -ENOTSUP;
2328 }
2329
2330 memset(res, 0, sizeof(*res));
2331 return bs->drv->bdrv_check(bs, res, fix);
2332 }
2333
2334 #define COMMIT_BUF_SECTORS 2048
2335
2336 /* commit COW file into the raw image */
2337 int bdrv_commit(BlockDriverState *bs)
2338 {
2339 BlockDriver *drv = bs->drv;
2340 int64_t sector, total_sectors, length, backing_length;
2341 int n, ro, open_flags;
2342 int ret = 0;
2343 uint8_t *buf = NULL;
2344
2345 if (!drv)
2346 return -ENOMEDIUM;
2347
2348 if (!bs->backing) {
2349 return -ENOTSUP;
2350 }
2351
2352 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, NULL) ||
2353 bdrv_op_is_blocked(bs->backing->bs, BLOCK_OP_TYPE_COMMIT_TARGET, NULL)) {
2354 return -EBUSY;
2355 }
2356
2357 ro = bs->backing->bs->read_only;
2358 open_flags = bs->backing->bs->open_flags;
2359
2360 if (ro) {
2361 if (bdrv_reopen(bs->backing->bs, open_flags | BDRV_O_RDWR, NULL)) {
2362 return -EACCES;
2363 }
2364 }
2365
2366 length = bdrv_getlength(bs);
2367 if (length < 0) {
2368 ret = length;
2369 goto ro_cleanup;
2370 }
2371
2372 backing_length = bdrv_getlength(bs->backing->bs);
2373 if (backing_length < 0) {
2374 ret = backing_length;
2375 goto ro_cleanup;
2376 }
2377
2378 /* If our top snapshot is larger than the backing file image,
2379 * grow the backing file image if possible. If not possible,
2380 * we must return an error */
2381 if (length > backing_length) {
2382 ret = bdrv_truncate(bs->backing->bs, length);
2383 if (ret < 0) {
2384 goto ro_cleanup;
2385 }
2386 }
2387
2388 total_sectors = length >> BDRV_SECTOR_BITS;
2389
2390 /* qemu_try_blockalign() for bs will choose an alignment that works for
2391 * bs->backing->bs as well, so no need to compare the alignment manually. */
2392 buf = qemu_try_blockalign(bs, COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2393 if (buf == NULL) {
2394 ret = -ENOMEM;
2395 goto ro_cleanup;
2396 }
2397
2398 for (sector = 0; sector < total_sectors; sector += n) {
2399 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2400 if (ret < 0) {
2401 goto ro_cleanup;
2402 }
2403 if (ret) {
2404 ret = bdrv_read(bs, sector, buf, n);
2405 if (ret < 0) {
2406 goto ro_cleanup;
2407 }
2408
2409 ret = bdrv_write(bs->backing->bs, sector, buf, n);
2410 if (ret < 0) {
2411 goto ro_cleanup;
2412 }
2413 }
2414 }
2415
2416 if (drv->bdrv_make_empty) {
2417 ret = drv->bdrv_make_empty(bs);
2418 if (ret < 0) {
2419 goto ro_cleanup;
2420 }
2421 bdrv_flush(bs);
2422 }
2423
2424 /*
2425 * Make sure all data we wrote to the backing device is actually
2426 * stable on disk.
2427 */
2428 if (bs->backing) {
2429 bdrv_flush(bs->backing->bs);
2430 }
2431
2432 ret = 0;
2433 ro_cleanup:
2434 qemu_vfree(buf);
2435
2436 if (ro) {
2437 /* ignoring error return here */
2438 bdrv_reopen(bs->backing->bs, open_flags & ~BDRV_O_RDWR, NULL);
2439 }
2440
2441 return ret;
2442 }
2443
2444 /*
2445 * Return values:
2446 * 0 - success
2447 * -EINVAL - backing format specified, but no file
2448 * -ENOSPC - can't update the backing file because no space is left in the
2449 * image file header
2450 * -ENOTSUP - format driver doesn't support changing the backing file
2451 */
2452 int bdrv_change_backing_file(BlockDriverState *bs,
2453 const char *backing_file, const char *backing_fmt)
2454 {
2455 BlockDriver *drv = bs->drv;
2456 int ret;
2457
2458 /* Backing file format doesn't make sense without a backing file */
2459 if (backing_fmt && !backing_file) {
2460 return -EINVAL;
2461 }
2462
2463 if (drv->bdrv_change_backing_file != NULL) {
2464 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2465 } else {
2466 ret = -ENOTSUP;
2467 }
2468
2469 if (ret == 0) {
2470 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2471 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2472 }
2473 return ret;
2474 }
2475
2476 /*
2477 * Finds the image layer in the chain that has 'bs' as its backing file.
2478 *
2479 * active is the current topmost image.
2480 *
2481 * Returns NULL if bs is not found in active's image chain,
2482 * or if active == bs.
2483 *
2484 * Returns the bottommost base image if bs == NULL.
2485 */
2486 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2487 BlockDriverState *bs)
2488 {
2489 while (active && bs != backing_bs(active)) {
2490 active = backing_bs(active);
2491 }
2492
2493 return active;
2494 }
2495
2496 /* Given a BDS, searches for the base layer. */
2497 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
2498 {
2499 return bdrv_find_overlay(bs, NULL);
2500 }
2501
2502 /*
2503 * Drops images above 'base' up to and including 'top', and sets the image
2504 * above 'top' to have base as its backing file.
2505 *
2506 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2507 * information in 'bs' can be properly updated.
2508 *
2509 * E.g., this will convert the following chain:
2510 * bottom <- base <- intermediate <- top <- active
2511 *
2512 * to
2513 *
2514 * bottom <- base <- active
2515 *
2516 * It is allowed for bottom==base, in which case it converts:
2517 *
2518 * base <- intermediate <- top <- active
2519 *
2520 * to
2521 *
2522 * base <- active
2523 *
2524 * If backing_file_str is non-NULL, it will be used when modifying top's
2525 * overlay image metadata.
2526 *
2527 * Error conditions:
2528 * if active == top, that is considered an error
2529 *
2530 */
2531 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2532 BlockDriverState *base, const char *backing_file_str)
2533 {
2534 BlockDriverState *new_top_bs = NULL;
2535 int ret = -EIO;
2536
2537 if (!top->drv || !base->drv) {
2538 goto exit;
2539 }
2540
2541 new_top_bs = bdrv_find_overlay(active, top);
2542
2543 if (new_top_bs == NULL) {
2544 /* we could not find the image above 'top', this is an error */
2545 goto exit;
2546 }
2547
2548 /* special case of new_top_bs->backing->bs already pointing to base - nothing
2549 * to do, no intermediate images */
2550 if (backing_bs(new_top_bs) == base) {
2551 ret = 0;
2552 goto exit;
2553 }
2554
2555 /* Make sure that base is in the backing chain of top */
2556 if (!bdrv_chain_contains(top, base)) {
2557 goto exit;
2558 }
2559
2560 /* success - we can delete the intermediate states, and link top->base */
2561 backing_file_str = backing_file_str ? backing_file_str : base->filename;
2562 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
2563 base->drv ? base->drv->format_name : "");
2564 if (ret) {
2565 goto exit;
2566 }
2567 bdrv_set_backing_hd(new_top_bs, base);
2568
2569 ret = 0;
2570 exit:
2571 return ret;
2572 }
2573
2574 /**
2575 * Truncate file to 'offset' bytes (needed only for file protocols)
2576 */
2577 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2578 {
2579 BlockDriver *drv = bs->drv;
2580 int ret;
2581 if (!drv)
2582 return -ENOMEDIUM;
2583 if (!drv->bdrv_truncate)
2584 return -ENOTSUP;
2585 if (bs->read_only)
2586 return -EACCES;
2587
2588 ret = drv->bdrv_truncate(bs, offset);
2589 if (ret == 0) {
2590 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2591 bdrv_dirty_bitmap_truncate(bs);
2592 bdrv_parent_cb_resize(bs);
2593 }
2594 return ret;
2595 }
2596
2597 /**
2598 * Length of a allocated file in bytes. Sparse files are counted by actual
2599 * allocated space. Return < 0 if error or unknown.
2600 */
2601 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2602 {
2603 BlockDriver *drv = bs->drv;
2604 if (!drv) {
2605 return -ENOMEDIUM;
2606 }
2607 if (drv->bdrv_get_allocated_file_size) {
2608 return drv->bdrv_get_allocated_file_size(bs);
2609 }
2610 if (bs->file) {
2611 return bdrv_get_allocated_file_size(bs->file->bs);
2612 }
2613 return -ENOTSUP;
2614 }
2615
2616 /**
2617 * Return number of sectors on success, -errno on error.
2618 */
2619 int64_t bdrv_nb_sectors(BlockDriverState *bs)
2620 {
2621 BlockDriver *drv = bs->drv;
2622
2623 if (!drv)
2624 return -ENOMEDIUM;
2625
2626 if (drv->has_variable_length) {
2627 int ret = refresh_total_sectors(bs, bs->total_sectors);
2628 if (ret < 0) {
2629 return ret;
2630 }
2631 }
2632 return bs->total_sectors;
2633 }
2634
2635 /**
2636 * Return length in bytes on success, -errno on error.
2637 * The length is always a multiple of BDRV_SECTOR_SIZE.
2638 */
2639 int64_t bdrv_getlength(BlockDriverState *bs)
2640 {
2641 int64_t ret = bdrv_nb_sectors(bs);
2642
2643 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
2644 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
2645 }
2646
2647 /* return 0 as number of sectors if no device present or error */
2648 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2649 {
2650 int64_t nb_sectors = bdrv_nb_sectors(bs);
2651
2652 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
2653 }
2654
2655 int bdrv_is_read_only(BlockDriverState *bs)
2656 {
2657 return bs->read_only;
2658 }
2659
2660 int bdrv_is_sg(BlockDriverState *bs)
2661 {
2662 return bs->sg;
2663 }
2664
2665 int bdrv_is_encrypted(BlockDriverState *bs)
2666 {
2667 if (bs->backing && bs->backing->bs->encrypted) {
2668 return 1;
2669 }
2670 return bs->encrypted;
2671 }
2672
2673 int bdrv_key_required(BlockDriverState *bs)
2674 {
2675 BdrvChild *backing = bs->backing;
2676
2677 if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
2678 return 1;
2679 }
2680 return (bs->encrypted && !bs->valid_key);
2681 }
2682
2683 int bdrv_set_key(BlockDriverState *bs, const char *key)
2684 {
2685 int ret;
2686 if (bs->backing && bs->backing->bs->encrypted) {
2687 ret = bdrv_set_key(bs->backing->bs, key);
2688 if (ret < 0)
2689 return ret;
2690 if (!bs->encrypted)
2691 return 0;
2692 }
2693 if (!bs->encrypted) {
2694 return -EINVAL;
2695 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2696 return -ENOMEDIUM;
2697 }
2698 ret = bs->drv->bdrv_set_key(bs, key);
2699 if (ret < 0) {
2700 bs->valid_key = 0;
2701 } else if (!bs->valid_key) {
2702 /* call the change callback now, we skipped it on open */
2703 bs->valid_key = 1;
2704 bdrv_parent_cb_change_media(bs, true);
2705 }
2706 return ret;
2707 }
2708
2709 /*
2710 * Provide an encryption key for @bs.
2711 * If @key is non-null:
2712 * If @bs is not encrypted, fail.
2713 * Else if the key is invalid, fail.
2714 * Else set @bs's key to @key, replacing the existing key, if any.
2715 * If @key is null:
2716 * If @bs is encrypted and still lacks a key, fail.
2717 * Else do nothing.
2718 * On failure, store an error object through @errp if non-null.
2719 */
2720 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
2721 {
2722 if (key) {
2723 if (!bdrv_is_encrypted(bs)) {
2724 error_setg(errp, "Node '%s' is not encrypted",
2725 bdrv_get_device_or_node_name(bs));
2726 } else if (bdrv_set_key(bs, key) < 0) {
2727 error_setg(errp, QERR_INVALID_PASSWORD);
2728 }
2729 } else {
2730 if (bdrv_key_required(bs)) {
2731 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
2732 "'%s' (%s) is encrypted",
2733 bdrv_get_device_or_node_name(bs),
2734 bdrv_get_encrypted_filename(bs));
2735 }
2736 }
2737 }
2738
2739 const char *bdrv_get_format_name(BlockDriverState *bs)
2740 {
2741 return bs->drv ? bs->drv->format_name : NULL;
2742 }
2743
2744 static int qsort_strcmp(const void *a, const void *b)
2745 {
2746 return strcmp(a, b);
2747 }
2748
2749 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2750 void *opaque)
2751 {
2752 BlockDriver *drv;
2753 int count = 0;
2754 int i;
2755 const char **formats = NULL;
2756
2757 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2758 if (drv->format_name) {
2759 bool found = false;
2760 int i = count;
2761 while (formats && i && !found) {
2762 found = !strcmp(formats[--i], drv->format_name);
2763 }
2764
2765 if (!found) {
2766 formats = g_renew(const char *, formats, count + 1);
2767 formats[count++] = drv->format_name;
2768 }
2769 }
2770 }
2771
2772 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
2773
2774 for (i = 0; i < count; i++) {
2775 it(opaque, formats[i]);
2776 }
2777
2778 g_free(formats);
2779 }
2780
2781 /* This function is to find a node in the bs graph */
2782 BlockDriverState *bdrv_find_node(const char *node_name)
2783 {
2784 BlockDriverState *bs;
2785
2786 assert(node_name);
2787
2788 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2789 if (!strcmp(node_name, bs->node_name)) {
2790 return bs;
2791 }
2792 }
2793 return NULL;
2794 }
2795
2796 /* Put this QMP function here so it can access the static graph_bdrv_states. */
2797 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
2798 {
2799 BlockDeviceInfoList *list, *entry;
2800 BlockDriverState *bs;
2801
2802 list = NULL;
2803 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2804 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
2805 if (!info) {
2806 qapi_free_BlockDeviceInfoList(list);
2807 return NULL;
2808 }
2809 entry = g_malloc0(sizeof(*entry));
2810 entry->value = info;
2811 entry->next = list;
2812 list = entry;
2813 }
2814
2815 return list;
2816 }
2817
2818 BlockDriverState *bdrv_lookup_bs(const char *device,
2819 const char *node_name,
2820 Error **errp)
2821 {
2822 BlockBackend *blk;
2823 BlockDriverState *bs;
2824
2825 if (device) {
2826 blk = blk_by_name(device);
2827
2828 if (blk) {
2829 bs = blk_bs(blk);
2830 if (!bs) {
2831 error_setg(errp, "Device '%s' has no medium", device);
2832 }
2833
2834 return bs;
2835 }
2836 }
2837
2838 if (node_name) {
2839 bs = bdrv_find_node(node_name);
2840
2841 if (bs) {
2842 return bs;
2843 }
2844 }
2845
2846 error_setg(errp, "Cannot find device=%s nor node_name=%s",
2847 device ? device : "",
2848 node_name ? node_name : "");
2849 return NULL;
2850 }
2851
2852 /* If 'base' is in the same chain as 'top', return true. Otherwise,
2853 * return false. If either argument is NULL, return false. */
2854 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
2855 {
2856 while (top && top != base) {
2857 top = backing_bs(top);
2858 }
2859
2860 return top != NULL;
2861 }
2862
2863 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
2864 {
2865 if (!bs) {
2866 return QTAILQ_FIRST(&graph_bdrv_states);
2867 }
2868 return QTAILQ_NEXT(bs, node_list);
2869 }
2870
2871 const char *bdrv_get_node_name(const BlockDriverState *bs)
2872 {
2873 return bs->node_name;
2874 }
2875
2876 const char *bdrv_get_parent_name(const BlockDriverState *bs)
2877 {
2878 BdrvChild *c;
2879 const char *name;
2880
2881 /* If multiple parents have a name, just pick the first one. */
2882 QLIST_FOREACH(c, &bs->parents, next_parent) {
2883 if (c->role->get_name) {
2884 name = c->role->get_name(c);
2885 if (name && *name) {
2886 return name;
2887 }
2888 }
2889 }
2890
2891 return NULL;
2892 }
2893
2894 /* TODO check what callers really want: bs->node_name or blk_name() */
2895 const char *bdrv_get_device_name(const BlockDriverState *bs)
2896 {
2897 return bdrv_get_parent_name(bs) ?: "";
2898 }
2899
2900 /* This can be used to identify nodes that might not have a device
2901 * name associated. Since node and device names live in the same
2902 * namespace, the result is unambiguous. The exception is if both are
2903 * absent, then this returns an empty (non-null) string. */
2904 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
2905 {
2906 return bdrv_get_parent_name(bs) ?: bs->node_name;
2907 }
2908
2909 int bdrv_get_flags(BlockDriverState *bs)
2910 {
2911 return bs->open_flags;
2912 }
2913
2914 int bdrv_has_zero_init_1(BlockDriverState *bs)
2915 {
2916 return 1;
2917 }
2918
2919 int bdrv_has_zero_init(BlockDriverState *bs)
2920 {
2921 assert(bs->drv);
2922
2923 /* If BS is a copy on write image, it is initialized to
2924 the contents of the base image, which may not be zeroes. */
2925 if (bs->backing) {
2926 return 0;
2927 }
2928 if (bs->drv->bdrv_has_zero_init) {
2929 return bs->drv->bdrv_has_zero_init(bs);
2930 }
2931
2932 /* safe default */
2933 return 0;
2934 }
2935
2936 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
2937 {
2938 BlockDriverInfo bdi;
2939
2940 if (bs->backing) {
2941 return false;
2942 }
2943
2944 if (bdrv_get_info(bs, &bdi) == 0) {
2945 return bdi.unallocated_blocks_are_zero;
2946 }
2947
2948 return false;
2949 }
2950
2951 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
2952 {
2953 BlockDriverInfo bdi;
2954
2955 if (bs->backing || !(bs->open_flags & BDRV_O_UNMAP)) {
2956 return false;
2957 }
2958
2959 if (bdrv_get_info(bs, &bdi) == 0) {
2960 return bdi.can_write_zeroes_with_unmap;
2961 }
2962
2963 return false;
2964 }
2965
2966 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2967 {
2968 if (bs->backing && bs->backing->bs->encrypted)
2969 return bs->backing_file;
2970 else if (bs->encrypted)
2971 return bs->filename;
2972 else
2973 return NULL;
2974 }
2975
2976 void bdrv_get_backing_filename(BlockDriverState *bs,
2977 char *filename, int filename_size)
2978 {
2979 pstrcpy(filename, filename_size, bs->backing_file);
2980 }
2981
2982 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2983 {
2984 BlockDriver *drv = bs->drv;
2985 if (!drv)
2986 return -ENOMEDIUM;
2987 if (!drv->bdrv_get_info)
2988 return -ENOTSUP;
2989 memset(bdi, 0, sizeof(*bdi));
2990 return drv->bdrv_get_info(bs, bdi);
2991 }
2992
2993 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
2994 {
2995 BlockDriver *drv = bs->drv;
2996 if (drv && drv->bdrv_get_specific_info) {
2997 return drv->bdrv_get_specific_info(bs);
2998 }
2999 return NULL;
3000 }
3001
3002 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3003 {
3004 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3005 return;
3006 }
3007
3008 bs->drv->bdrv_debug_event(bs, event);
3009 }
3010
3011 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3012 const char *tag)
3013 {
3014 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3015 bs = bs->file ? bs->file->bs : NULL;
3016 }
3017
3018 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3019 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3020 }
3021
3022 return -ENOTSUP;
3023 }
3024
3025 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3026 {
3027 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3028 bs = bs->file ? bs->file->bs : NULL;
3029 }
3030
3031 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3032 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3033 }
3034
3035 return -ENOTSUP;
3036 }
3037
3038 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3039 {
3040 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3041 bs = bs->file ? bs->file->bs : NULL;
3042 }
3043
3044 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3045 return bs->drv->bdrv_debug_resume(bs, tag);
3046 }
3047
3048 return -ENOTSUP;
3049 }
3050
3051 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3052 {
3053 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3054 bs = bs->file ? bs->file->bs : NULL;
3055 }
3056
3057 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3058 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3059 }
3060
3061 return false;
3062 }
3063
3064 int bdrv_is_snapshot(BlockDriverState *bs)
3065 {
3066 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
3067 }
3068
3069 /* backing_file can either be relative, or absolute, or a protocol. If it is
3070 * relative, it must be relative to the chain. So, passing in bs->filename
3071 * from a BDS as backing_file should not be done, as that may be relative to
3072 * the CWD rather than the chain. */
3073 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3074 const char *backing_file)
3075 {
3076 char *filename_full = NULL;
3077 char *backing_file_full = NULL;
3078 char *filename_tmp = NULL;
3079 int is_protocol = 0;
3080 BlockDriverState *curr_bs = NULL;
3081 BlockDriverState *retval = NULL;
3082
3083 if (!bs || !bs->drv || !backing_file) {
3084 return NULL;
3085 }
3086
3087 filename_full = g_malloc(PATH_MAX);
3088 backing_file_full = g_malloc(PATH_MAX);
3089 filename_tmp = g_malloc(PATH_MAX);
3090
3091 is_protocol = path_has_protocol(backing_file);
3092
3093 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3094
3095 /* If either of the filename paths is actually a protocol, then
3096 * compare unmodified paths; otherwise make paths relative */
3097 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3098 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3099 retval = curr_bs->backing->bs;
3100 break;
3101 }
3102 } else {
3103 /* If not an absolute filename path, make it relative to the current
3104 * image's filename path */
3105 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3106 backing_file);
3107
3108 /* We are going to compare absolute pathnames */
3109 if (!realpath(filename_tmp, filename_full)) {
3110 continue;
3111 }
3112
3113 /* We need to make sure the backing filename we are comparing against
3114 * is relative to the current image filename (or absolute) */
3115 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3116 curr_bs->backing_file);
3117
3118 if (!realpath(filename_tmp, backing_file_full)) {
3119 continue;
3120 }
3121
3122 if (strcmp(backing_file_full, filename_full) == 0) {
3123 retval = curr_bs->backing->bs;
3124 break;
3125 }
3126 }
3127 }
3128
3129 g_free(filename_full);
3130 g_free(backing_file_full);
3131 g_free(filename_tmp);
3132 return retval;
3133 }
3134
3135 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3136 {
3137 if (!bs->drv) {
3138 return 0;
3139 }
3140
3141 if (!bs->backing) {
3142 return 0;
3143 }
3144
3145 return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3146 }
3147
3148 void bdrv_init(void)
3149 {
3150 module_call_init(MODULE_INIT_BLOCK);
3151 }
3152
3153 void bdrv_init_with_whitelist(void)
3154 {
3155 use_bdrv_whitelist = 1;
3156 bdrv_init();
3157 }
3158
3159 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3160 {
3161 BdrvChild *child;
3162 Error *local_err = NULL;
3163 int ret;
3164
3165 if (!bs->drv) {
3166 return;
3167 }
3168
3169 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3170 return;
3171 }
3172 bs->open_flags &= ~BDRV_O_INACTIVE;
3173
3174 if (bs->drv->bdrv_invalidate_cache) {
3175 bs->drv->bdrv_invalidate_cache(bs, &local_err);
3176 if (local_err) {
3177 bs->open_flags |= BDRV_O_INACTIVE;
3178 error_propagate(errp, local_err);
3179 return;
3180 }
3181 }
3182
3183 QLIST_FOREACH(child, &bs->children, next) {
3184 bdrv_invalidate_cache(child->bs, &local_err);
3185 if (local_err) {
3186 bs->open_flags |= BDRV_O_INACTIVE;
3187 error_propagate(errp, local_err);
3188 return;
3189 }
3190 }
3191
3192 ret = refresh_total_sectors(bs, bs->total_sectors);
3193 if (ret < 0) {
3194 bs->open_flags |= BDRV_O_INACTIVE;
3195 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3196 return;
3197 }
3198 }
3199
3200 void bdrv_invalidate_cache_all(Error **errp)
3201 {
3202 BlockDriverState *bs;
3203 Error *local_err = NULL;
3204 BdrvNextIterator it;
3205
3206 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3207 AioContext *aio_context = bdrv_get_aio_context(bs);
3208
3209 aio_context_acquire(aio_context);
3210 bdrv_invalidate_cache(bs, &local_err);
3211 aio_context_release(aio_context);
3212 if (local_err) {
3213 error_propagate(errp, local_err);
3214 return;
3215 }
3216 }
3217 }
3218
3219 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3220 bool setting_flag)
3221 {
3222 BdrvChild *child;
3223 int ret;
3224
3225 if (!setting_flag && bs->drv->bdrv_inactivate) {
3226 ret = bs->drv->bdrv_inactivate(bs);
3227 if (ret < 0) {
3228 return ret;
3229 }
3230 }
3231
3232 QLIST_FOREACH(child, &bs->children, next) {
3233 ret = bdrv_inactivate_recurse(child->bs, setting_flag);
3234 if (ret < 0) {
3235 return ret;
3236 }
3237 }
3238
3239 if (setting_flag) {
3240 bs->open_flags |= BDRV_O_INACTIVE;
3241 }
3242 return 0;
3243 }
3244
3245 int bdrv_inactivate_all(void)
3246 {
3247 BlockDriverState *bs = NULL;
3248 BdrvNextIterator it;
3249 int ret = 0;
3250 int pass;
3251
3252 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3253 aio_context_acquire(bdrv_get_aio_context(bs));
3254 }
3255
3256 /* We do two passes of inactivation. The first pass calls to drivers'
3257 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
3258 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
3259 * is allowed. */
3260 for (pass = 0; pass < 2; pass++) {
3261 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3262 ret = bdrv_inactivate_recurse(bs, pass);
3263 if (ret < 0) {
3264 goto out;
3265 }
3266 }
3267 }
3268
3269 out:
3270 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3271 aio_context_release(bdrv_get_aio_context(bs));
3272 }
3273
3274 return ret;
3275 }
3276
3277 /**************************************************************/
3278 /* removable device support */
3279
3280 /**
3281 * Return TRUE if the media is present
3282 */
3283 bool bdrv_is_inserted(BlockDriverState *bs)
3284 {
3285 BlockDriver *drv = bs->drv;
3286 BdrvChild *child;
3287
3288 if (!drv) {
3289 return false;
3290 }
3291 if (drv->bdrv_is_inserted) {
3292 return drv->bdrv_is_inserted(bs);
3293 }
3294 QLIST_FOREACH(child, &bs->children, next) {
3295 if (!bdrv_is_inserted(child->bs)) {
3296 return false;
3297 }
3298 }
3299 return true;
3300 }
3301
3302 /**
3303 * Return whether the media changed since the last call to this
3304 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3305 */
3306 int bdrv_media_changed(BlockDriverState *bs)
3307 {
3308 BlockDriver *drv = bs->drv;
3309
3310 if (drv && drv->bdrv_media_changed) {
3311 return drv->bdrv_media_changed(bs);
3312 }
3313 return -ENOTSUP;
3314 }
3315
3316 /**
3317 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3318 */
3319 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3320 {
3321 BlockDriver *drv = bs->drv;
3322 const char *device_name;
3323
3324 if (drv && drv->bdrv_eject) {
3325 drv->bdrv_eject(bs, eject_flag);
3326 }
3327
3328 device_name = bdrv_get_device_name(bs);
3329 if (device_name[0] != '\0') {
3330 qapi_event_send_device_tray_moved(device_name,
3331 eject_flag, &error_abort);
3332 }
3333 }
3334
3335 /**
3336 * Lock or unlock the media (if it is locked, the user won't be able
3337 * to eject it manually).
3338 */
3339 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3340 {
3341 BlockDriver *drv = bs->drv;
3342
3343 trace_bdrv_lock_medium(bs, locked);
3344
3345 if (drv && drv->bdrv_lock_medium) {
3346 drv->bdrv_lock_medium(bs, locked);
3347 }
3348 }
3349
3350 /* Get a reference to bs */
3351 void bdrv_ref(BlockDriverState *bs)
3352 {
3353 bs->refcnt++;
3354 }
3355
3356 /* Release a previously grabbed reference to bs.
3357 * If after releasing, reference count is zero, the BlockDriverState is
3358 * deleted. */
3359 void bdrv_unref(BlockDriverState *bs)
3360 {
3361 if (!bs) {
3362 return;
3363 }
3364 assert(bs->refcnt > 0);
3365 if (--bs->refcnt == 0) {
3366 bdrv_delete(bs);
3367 }
3368 }
3369
3370 struct BdrvOpBlocker {
3371 Error *reason;
3372 QLIST_ENTRY(BdrvOpBlocker) list;
3373 };
3374
3375 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
3376 {
3377 BdrvOpBlocker *blocker;
3378 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3379 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
3380 blocker = QLIST_FIRST(&bs->op_blockers[op]);
3381 if (errp) {
3382 *errp = error_copy(blocker->reason);
3383 error_prepend(errp, "Node '%s' is busy: ",
3384 bdrv_get_device_or_node_name(bs));
3385 }
3386 return true;
3387 }
3388 return false;
3389 }
3390
3391 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
3392 {
3393 BdrvOpBlocker *blocker;
3394 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3395
3396 blocker = g_new0(BdrvOpBlocker, 1);
3397 blocker->reason = reason;
3398 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
3399 }
3400
3401 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
3402 {
3403 BdrvOpBlocker *blocker, *next;
3404 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3405 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
3406 if (blocker->reason == reason) {
3407 QLIST_REMOVE(blocker, list);
3408 g_free(blocker);
3409 }
3410 }
3411 }
3412
3413 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
3414 {
3415 int i;
3416 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3417 bdrv_op_block(bs, i, reason);
3418 }
3419 }
3420
3421 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
3422 {
3423 int i;
3424 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3425 bdrv_op_unblock(bs, i, reason);
3426 }
3427 }
3428
3429 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
3430 {
3431 int i;
3432
3433 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3434 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
3435 return false;
3436 }
3437 }
3438 return true;
3439 }
3440
3441 void bdrv_img_create(const char *filename, const char *fmt,
3442 const char *base_filename, const char *base_fmt,
3443 char *options, uint64_t img_size, int flags,
3444 Error **errp, bool quiet)
3445 {
3446 QemuOptsList *create_opts = NULL;
3447 QemuOpts *opts = NULL;
3448 const char *backing_fmt, *backing_file;
3449 int64_t size;
3450 BlockDriver *drv, *proto_drv;
3451 Error *local_err = NULL;
3452 int ret = 0;
3453
3454 /* Find driver and parse its options */
3455 drv = bdrv_find_format(fmt);
3456 if (!drv) {
3457 error_setg(errp, "Unknown file format '%s'", fmt);
3458 return;
3459 }
3460
3461 proto_drv = bdrv_find_protocol(filename, true, errp);
3462 if (!proto_drv) {
3463 return;
3464 }
3465
3466 if (!drv->create_opts) {
3467 error_setg(errp, "Format driver '%s' does not support image creation",
3468 drv->format_name);
3469 return;
3470 }
3471
3472 if (!proto_drv->create_opts) {
3473 error_setg(errp, "Protocol driver '%s' does not support image creation",
3474 proto_drv->format_name);
3475 return;
3476 }
3477
3478 create_opts = qemu_opts_append(create_opts, drv->create_opts);
3479 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
3480
3481 /* Create parameter list with default values */
3482 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
3483 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
3484
3485 /* Parse -o options */
3486 if (options) {
3487 qemu_opts_do_parse(opts, options, NULL, &local_err);
3488 if (local_err) {
3489 error_report_err(local_err);
3490 local_err = NULL;
3491 error_setg(errp, "Invalid options for file format '%s'", fmt);
3492 goto out;
3493 }
3494 }
3495
3496 if (base_filename) {
3497 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
3498 if (local_err) {
3499 error_setg(errp, "Backing file not supported for file format '%s'",
3500 fmt);
3501 goto out;
3502 }
3503 }
3504
3505 if (base_fmt) {
3506 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
3507 if (local_err) {
3508 error_setg(errp, "Backing file format not supported for file "
3509 "format '%s'", fmt);
3510 goto out;
3511 }
3512 }
3513
3514 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3515 if (backing_file) {
3516 if (!strcmp(filename, backing_file)) {
3517 error_setg(errp, "Error: Trying to create an image with the "
3518 "same filename as the backing file");
3519 goto out;
3520 }
3521 }
3522
3523 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3524
3525 // The size for the image must always be specified, with one exception:
3526 // If we are using a backing file, we can obtain the size from there
3527 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3528 if (size == -1) {
3529 if (backing_file) {
3530 BlockDriverState *bs;
3531 char *full_backing = g_new0(char, PATH_MAX);
3532 int64_t size;
3533 int back_flags;
3534 QDict *backing_options = NULL;
3535
3536 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
3537 full_backing, PATH_MAX,
3538 &local_err);
3539 if (local_err) {
3540 g_free(full_backing);
3541 goto out;
3542 }
3543
3544 /* backing files always opened read-only */
3545 back_flags = flags;
3546 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
3547
3548 if (backing_fmt) {
3549 backing_options = qdict_new();
3550 qdict_put(backing_options, "driver",
3551 qstring_from_str(backing_fmt));
3552 }
3553
3554 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
3555 &local_err);
3556 g_free(full_backing);
3557 if (!bs) {
3558 goto out;
3559 }
3560 size = bdrv_getlength(bs);
3561 if (size < 0) {
3562 error_setg_errno(errp, -size, "Could not get size of '%s'",
3563 backing_file);
3564 bdrv_unref(bs);
3565 goto out;
3566 }
3567
3568 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
3569
3570 bdrv_unref(bs);
3571 } else {
3572 error_setg(errp, "Image creation needs a size parameter");
3573 goto out;
3574 }
3575 }
3576
3577 if (!quiet) {
3578 printf("Formatting '%s', fmt=%s ", filename, fmt);
3579 qemu_opts_print(opts, " ");
3580 puts("");
3581 }
3582
3583 ret = bdrv_create<