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