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