qcow2: Fix order in L2 table COW
[qemu/cris-port.git] / block.c
blobb476479e0dc9c1cb1f659c4a5db1af9e3e1005e3
1 /*
2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "config-host.h"
25 #include "qemu-common.h"
26 #include "trace.h"
27 #include "monitor.h"
28 #include "block_int.h"
29 #include "module.h"
30 #include "qemu-objects.h"
32 #ifdef CONFIG_BSD
33 #include <sys/types.h>
34 #include <sys/stat.h>
35 #include <sys/ioctl.h>
36 #include <sys/queue.h>
37 #ifndef __DragonFly__
38 #include <sys/disk.h>
39 #endif
40 #endif
42 #ifdef _WIN32
43 #include <windows.h>
44 #endif
46 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
47 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
48 BlockDriverCompletionFunc *cb, void *opaque);
49 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
50 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
51 BlockDriverCompletionFunc *cb, void *opaque);
52 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
53 BlockDriverCompletionFunc *cb, void *opaque);
54 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
55 BlockDriverCompletionFunc *cb, void *opaque);
56 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
57 uint8_t *buf, int nb_sectors);
58 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
59 const uint8_t *buf, int nb_sectors);
61 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(bdrv_states);
64 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
65 QLIST_HEAD_INITIALIZER(bdrv_drivers);
67 /* The device to use for VM snapshots */
68 static BlockDriverState *bs_snapshots;
70 /* If non-zero, use only whitelisted block drivers */
71 static int use_bdrv_whitelist;
73 #ifdef _WIN32
74 static int is_windows_drive_prefix(const char *filename)
76 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
77 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
78 filename[1] == ':');
81 int is_windows_drive(const char *filename)
83 if (is_windows_drive_prefix(filename) &&
84 filename[2] == '\0')
85 return 1;
86 if (strstart(filename, "\\\\.\\", NULL) ||
87 strstart(filename, "//./", NULL))
88 return 1;
89 return 0;
91 #endif
93 /* check if the path starts with "<protocol>:" */
94 static int path_has_protocol(const char *path)
96 #ifdef _WIN32
97 if (is_windows_drive(path) ||
98 is_windows_drive_prefix(path)) {
99 return 0;
101 #endif
103 return strchr(path, ':') != NULL;
106 int path_is_absolute(const char *path)
108 const char *p;
109 #ifdef _WIN32
110 /* specific case for names like: "\\.\d:" */
111 if (*path == '/' || *path == '\\')
112 return 1;
113 #endif
114 p = strchr(path, ':');
115 if (p)
116 p++;
117 else
118 p = path;
119 #ifdef _WIN32
120 return (*p == '/' || *p == '\\');
121 #else
122 return (*p == '/');
123 #endif
126 /* if filename is absolute, just copy it to dest. Otherwise, build a
127 path to it by considering it is relative to base_path. URL are
128 supported. */
129 void path_combine(char *dest, int dest_size,
130 const char *base_path,
131 const char *filename)
133 const char *p, *p1;
134 int len;
136 if (dest_size <= 0)
137 return;
138 if (path_is_absolute(filename)) {
139 pstrcpy(dest, dest_size, filename);
140 } else {
141 p = strchr(base_path, ':');
142 if (p)
143 p++;
144 else
145 p = base_path;
146 p1 = strrchr(base_path, '/');
147 #ifdef _WIN32
149 const char *p2;
150 p2 = strrchr(base_path, '\\');
151 if (!p1 || p2 > p1)
152 p1 = p2;
154 #endif
155 if (p1)
156 p1++;
157 else
158 p1 = base_path;
159 if (p1 > p)
160 p = p1;
161 len = p - base_path;
162 if (len > dest_size - 1)
163 len = dest_size - 1;
164 memcpy(dest, base_path, len);
165 dest[len] = '\0';
166 pstrcat(dest, dest_size, filename);
170 void bdrv_register(BlockDriver *bdrv)
172 if (!bdrv->bdrv_aio_readv) {
173 /* add AIO emulation layer */
174 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
175 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
176 } else if (!bdrv->bdrv_read) {
177 /* add synchronous IO emulation layer */
178 bdrv->bdrv_read = bdrv_read_em;
179 bdrv->bdrv_write = bdrv_write_em;
182 if (!bdrv->bdrv_aio_flush)
183 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
185 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
188 /* create a new block device (by default it is empty) */
189 BlockDriverState *bdrv_new(const char *device_name)
191 BlockDriverState *bs;
193 bs = qemu_mallocz(sizeof(BlockDriverState));
194 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
195 if (device_name[0] != '\0') {
196 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
198 return bs;
201 BlockDriver *bdrv_find_format(const char *format_name)
203 BlockDriver *drv1;
204 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
205 if (!strcmp(drv1->format_name, format_name)) {
206 return drv1;
209 return NULL;
212 static int bdrv_is_whitelisted(BlockDriver *drv)
214 static const char *whitelist[] = {
215 CONFIG_BDRV_WHITELIST
217 const char **p;
219 if (!whitelist[0])
220 return 1; /* no whitelist, anything goes */
222 for (p = whitelist; *p; p++) {
223 if (!strcmp(drv->format_name, *p)) {
224 return 1;
227 return 0;
230 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
232 BlockDriver *drv = bdrv_find_format(format_name);
233 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
236 int bdrv_create(BlockDriver *drv, const char* filename,
237 QEMUOptionParameter *options)
239 if (!drv->bdrv_create)
240 return -ENOTSUP;
242 return drv->bdrv_create(filename, options);
245 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
247 BlockDriver *drv;
249 drv = bdrv_find_protocol(filename);
250 if (drv == NULL) {
251 return -ENOENT;
254 return bdrv_create(drv, filename, options);
257 #ifdef _WIN32
258 void get_tmp_filename(char *filename, int size)
260 char temp_dir[MAX_PATH];
262 GetTempPath(MAX_PATH, temp_dir);
263 GetTempFileName(temp_dir, "qem", 0, filename);
265 #else
266 void get_tmp_filename(char *filename, int size)
268 int fd;
269 const char *tmpdir;
270 /* XXX: race condition possible */
271 tmpdir = getenv("TMPDIR");
272 if (!tmpdir)
273 tmpdir = "/tmp";
274 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
275 fd = mkstemp(filename);
276 close(fd);
278 #endif
281 * Detect host devices. By convention, /dev/cdrom[N] is always
282 * recognized as a host CDROM.
284 static BlockDriver *find_hdev_driver(const char *filename)
286 int score_max = 0, score;
287 BlockDriver *drv = NULL, *d;
289 QLIST_FOREACH(d, &bdrv_drivers, list) {
290 if (d->bdrv_probe_device) {
291 score = d->bdrv_probe_device(filename);
292 if (score > score_max) {
293 score_max = score;
294 drv = d;
299 return drv;
302 BlockDriver *bdrv_find_protocol(const char *filename)
304 BlockDriver *drv1;
305 char protocol[128];
306 int len;
307 const char *p;
309 /* TODO Drivers without bdrv_file_open must be specified explicitly */
312 * XXX(hch): we really should not let host device detection
313 * override an explicit protocol specification, but moving this
314 * later breaks access to device names with colons in them.
315 * Thanks to the brain-dead persistent naming schemes on udev-
316 * based Linux systems those actually are quite common.
318 drv1 = find_hdev_driver(filename);
319 if (drv1) {
320 return drv1;
323 if (!path_has_protocol(filename)) {
324 return bdrv_find_format("file");
326 p = strchr(filename, ':');
327 assert(p != NULL);
328 len = p - filename;
329 if (len > sizeof(protocol) - 1)
330 len = sizeof(protocol) - 1;
331 memcpy(protocol, filename, len);
332 protocol[len] = '\0';
333 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
334 if (drv1->protocol_name &&
335 !strcmp(drv1->protocol_name, protocol)) {
336 return drv1;
339 return NULL;
342 static int find_image_format(const char *filename, BlockDriver **pdrv)
344 int ret, score, score_max;
345 BlockDriver *drv1, *drv;
346 uint8_t buf[2048];
347 BlockDriverState *bs;
349 ret = bdrv_file_open(&bs, filename, 0);
350 if (ret < 0) {
351 *pdrv = NULL;
352 return ret;
355 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
356 if (bs->sg || !bdrv_is_inserted(bs)) {
357 bdrv_delete(bs);
358 drv = bdrv_find_format("raw");
359 if (!drv) {
360 ret = -ENOENT;
362 *pdrv = drv;
363 return ret;
366 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
367 bdrv_delete(bs);
368 if (ret < 0) {
369 *pdrv = NULL;
370 return ret;
373 score_max = 0;
374 drv = NULL;
375 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
376 if (drv1->bdrv_probe) {
377 score = drv1->bdrv_probe(buf, ret, filename);
378 if (score > score_max) {
379 score_max = score;
380 drv = drv1;
384 if (!drv) {
385 ret = -ENOENT;
387 *pdrv = drv;
388 return ret;
392 * Set the current 'total_sectors' value
394 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
396 BlockDriver *drv = bs->drv;
398 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
399 if (bs->sg)
400 return 0;
402 /* query actual device if possible, otherwise just trust the hint */
403 if (drv->bdrv_getlength) {
404 int64_t length = drv->bdrv_getlength(bs);
405 if (length < 0) {
406 return length;
408 hint = length >> BDRV_SECTOR_BITS;
411 bs->total_sectors = hint;
412 return 0;
416 * Common part for opening disk images and files
418 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
419 int flags, BlockDriver *drv)
421 int ret, open_flags;
423 assert(drv != NULL);
425 bs->file = NULL;
426 bs->total_sectors = 0;
427 bs->encrypted = 0;
428 bs->valid_key = 0;
429 bs->open_flags = flags;
430 /* buffer_alignment defaulted to 512, drivers can change this value */
431 bs->buffer_alignment = 512;
433 pstrcpy(bs->filename, sizeof(bs->filename), filename);
435 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
436 return -ENOTSUP;
439 bs->drv = drv;
440 bs->opaque = qemu_mallocz(drv->instance_size);
443 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
444 * write cache to the guest. We do need the fdatasync to flush
445 * out transactions for block allocations, and we maybe have a
446 * volatile write cache in our backing device to deal with.
448 if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
449 bs->enable_write_cache = 1;
452 * Clear flags that are internal to the block layer before opening the
453 * image.
455 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
458 * Snapshots should be writeable.
460 if (bs->is_temporary) {
461 open_flags |= BDRV_O_RDWR;
464 /* Open the image, either directly or using a protocol */
465 if (drv->bdrv_file_open) {
466 ret = drv->bdrv_file_open(bs, filename, open_flags);
467 } else {
468 ret = bdrv_file_open(&bs->file, filename, open_flags);
469 if (ret >= 0) {
470 ret = drv->bdrv_open(bs, open_flags);
474 if (ret < 0) {
475 goto free_and_fail;
478 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
480 ret = refresh_total_sectors(bs, bs->total_sectors);
481 if (ret < 0) {
482 goto free_and_fail;
485 #ifndef _WIN32
486 if (bs->is_temporary) {
487 unlink(filename);
489 #endif
490 return 0;
492 free_and_fail:
493 if (bs->file) {
494 bdrv_delete(bs->file);
495 bs->file = NULL;
497 qemu_free(bs->opaque);
498 bs->opaque = NULL;
499 bs->drv = NULL;
500 return ret;
504 * Opens a file using a protocol (file, host_device, nbd, ...)
506 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
508 BlockDriverState *bs;
509 BlockDriver *drv;
510 int ret;
512 drv = bdrv_find_protocol(filename);
513 if (!drv) {
514 return -ENOENT;
517 bs = bdrv_new("");
518 ret = bdrv_open_common(bs, filename, flags, drv);
519 if (ret < 0) {
520 bdrv_delete(bs);
521 return ret;
523 bs->growable = 1;
524 *pbs = bs;
525 return 0;
529 * Opens a disk image (raw, qcow2, vmdk, ...)
531 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
532 BlockDriver *drv)
534 int ret;
536 if (flags & BDRV_O_SNAPSHOT) {
537 BlockDriverState *bs1;
538 int64_t total_size;
539 int is_protocol = 0;
540 BlockDriver *bdrv_qcow2;
541 QEMUOptionParameter *options;
542 char tmp_filename[PATH_MAX];
543 char backing_filename[PATH_MAX];
545 /* if snapshot, we create a temporary backing file and open it
546 instead of opening 'filename' directly */
548 /* if there is a backing file, use it */
549 bs1 = bdrv_new("");
550 ret = bdrv_open(bs1, filename, 0, drv);
551 if (ret < 0) {
552 bdrv_delete(bs1);
553 return ret;
555 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
557 if (bs1->drv && bs1->drv->protocol_name)
558 is_protocol = 1;
560 bdrv_delete(bs1);
562 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
564 /* Real path is meaningless for protocols */
565 if (is_protocol)
566 snprintf(backing_filename, sizeof(backing_filename),
567 "%s", filename);
568 else if (!realpath(filename, backing_filename))
569 return -errno;
571 bdrv_qcow2 = bdrv_find_format("qcow2");
572 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
574 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
575 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
576 if (drv) {
577 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
578 drv->format_name);
581 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
582 free_option_parameters(options);
583 if (ret < 0) {
584 return ret;
587 filename = tmp_filename;
588 drv = bdrv_qcow2;
589 bs->is_temporary = 1;
592 /* Find the right image format driver */
593 if (!drv) {
594 ret = find_image_format(filename, &drv);
597 if (!drv) {
598 goto unlink_and_fail;
601 /* Open the image */
602 ret = bdrv_open_common(bs, filename, flags, drv);
603 if (ret < 0) {
604 goto unlink_and_fail;
607 /* If there is a backing file, use it */
608 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
609 char backing_filename[PATH_MAX];
610 int back_flags;
611 BlockDriver *back_drv = NULL;
613 bs->backing_hd = bdrv_new("");
615 if (path_has_protocol(bs->backing_file)) {
616 pstrcpy(backing_filename, sizeof(backing_filename),
617 bs->backing_file);
618 } else {
619 path_combine(backing_filename, sizeof(backing_filename),
620 filename, bs->backing_file);
623 if (bs->backing_format[0] != '\0') {
624 back_drv = bdrv_find_format(bs->backing_format);
627 /* backing files always opened read-only */
628 back_flags =
629 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
631 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
632 if (ret < 0) {
633 bdrv_close(bs);
634 return ret;
636 if (bs->is_temporary) {
637 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
638 } else {
639 /* base image inherits from "parent" */
640 bs->backing_hd->keep_read_only = bs->keep_read_only;
644 if (!bdrv_key_required(bs)) {
645 /* call the change callback */
646 bs->media_changed = 1;
647 if (bs->change_cb)
648 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
651 return 0;
653 unlink_and_fail:
654 if (bs->is_temporary) {
655 unlink(filename);
657 return ret;
660 void bdrv_close(BlockDriverState *bs)
662 if (bs->drv) {
663 if (bs == bs_snapshots) {
664 bs_snapshots = NULL;
666 if (bs->backing_hd) {
667 bdrv_delete(bs->backing_hd);
668 bs->backing_hd = NULL;
670 bs->drv->bdrv_close(bs);
671 qemu_free(bs->opaque);
672 #ifdef _WIN32
673 if (bs->is_temporary) {
674 unlink(bs->filename);
676 #endif
677 bs->opaque = NULL;
678 bs->drv = NULL;
680 if (bs->file != NULL) {
681 bdrv_close(bs->file);
684 /* call the change callback */
685 bs->media_changed = 1;
686 if (bs->change_cb)
687 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
691 void bdrv_close_all(void)
693 BlockDriverState *bs;
695 QTAILQ_FOREACH(bs, &bdrv_states, list) {
696 bdrv_close(bs);
700 void bdrv_delete(BlockDriverState *bs)
702 assert(!bs->peer);
704 /* remove from list, if necessary */
705 if (bs->device_name[0] != '\0') {
706 QTAILQ_REMOVE(&bdrv_states, bs, list);
709 bdrv_close(bs);
710 if (bs->file != NULL) {
711 bdrv_delete(bs->file);
714 assert(bs != bs_snapshots);
715 qemu_free(bs);
718 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
720 if (bs->peer) {
721 return -EBUSY;
723 bs->peer = qdev;
724 return 0;
727 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
729 assert(bs->peer == qdev);
730 bs->peer = NULL;
733 DeviceState *bdrv_get_attached(BlockDriverState *bs)
735 return bs->peer;
739 * Run consistency checks on an image
741 * Returns 0 if the check could be completed (it doesn't mean that the image is
742 * free of errors) or -errno when an internal error occured. The results of the
743 * check are stored in res.
745 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
747 if (bs->drv->bdrv_check == NULL) {
748 return -ENOTSUP;
751 memset(res, 0, sizeof(*res));
752 return bs->drv->bdrv_check(bs, res);
755 #define COMMIT_BUF_SECTORS 2048
757 /* commit COW file into the raw image */
758 int bdrv_commit(BlockDriverState *bs)
760 BlockDriver *drv = bs->drv;
761 BlockDriver *backing_drv;
762 int64_t sector, total_sectors;
763 int n, ro, open_flags;
764 int ret = 0, rw_ret = 0;
765 uint8_t *buf;
766 char filename[1024];
767 BlockDriverState *bs_rw, *bs_ro;
769 if (!drv)
770 return -ENOMEDIUM;
772 if (!bs->backing_hd) {
773 return -ENOTSUP;
776 if (bs->backing_hd->keep_read_only) {
777 return -EACCES;
780 backing_drv = bs->backing_hd->drv;
781 ro = bs->backing_hd->read_only;
782 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
783 open_flags = bs->backing_hd->open_flags;
785 if (ro) {
786 /* re-open as RW */
787 bdrv_delete(bs->backing_hd);
788 bs->backing_hd = NULL;
789 bs_rw = bdrv_new("");
790 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
791 backing_drv);
792 if (rw_ret < 0) {
793 bdrv_delete(bs_rw);
794 /* try to re-open read-only */
795 bs_ro = bdrv_new("");
796 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
797 backing_drv);
798 if (ret < 0) {
799 bdrv_delete(bs_ro);
800 /* drive not functional anymore */
801 bs->drv = NULL;
802 return ret;
804 bs->backing_hd = bs_ro;
805 return rw_ret;
807 bs->backing_hd = bs_rw;
810 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
811 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
813 for (sector = 0; sector < total_sectors; sector += n) {
814 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
816 if (bdrv_read(bs, sector, buf, n) != 0) {
817 ret = -EIO;
818 goto ro_cleanup;
821 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
822 ret = -EIO;
823 goto ro_cleanup;
828 if (drv->bdrv_make_empty) {
829 ret = drv->bdrv_make_empty(bs);
830 bdrv_flush(bs);
834 * Make sure all data we wrote to the backing device is actually
835 * stable on disk.
837 if (bs->backing_hd)
838 bdrv_flush(bs->backing_hd);
840 ro_cleanup:
841 qemu_free(buf);
843 if (ro) {
844 /* re-open as RO */
845 bdrv_delete(bs->backing_hd);
846 bs->backing_hd = NULL;
847 bs_ro = bdrv_new("");
848 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
849 backing_drv);
850 if (ret < 0) {
851 bdrv_delete(bs_ro);
852 /* drive not functional anymore */
853 bs->drv = NULL;
854 return ret;
856 bs->backing_hd = bs_ro;
857 bs->backing_hd->keep_read_only = 0;
860 return ret;
863 void bdrv_commit_all(void)
865 BlockDriverState *bs;
867 QTAILQ_FOREACH(bs, &bdrv_states, list) {
868 bdrv_commit(bs);
873 * Return values:
874 * 0 - success
875 * -EINVAL - backing format specified, but no file
876 * -ENOSPC - can't update the backing file because no space is left in the
877 * image file header
878 * -ENOTSUP - format driver doesn't support changing the backing file
880 int bdrv_change_backing_file(BlockDriverState *bs,
881 const char *backing_file, const char *backing_fmt)
883 BlockDriver *drv = bs->drv;
885 if (drv->bdrv_change_backing_file != NULL) {
886 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
887 } else {
888 return -ENOTSUP;
892 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
893 size_t size)
895 int64_t len;
897 if (!bdrv_is_inserted(bs))
898 return -ENOMEDIUM;
900 if (bs->growable)
901 return 0;
903 len = bdrv_getlength(bs);
905 if (offset < 0)
906 return -EIO;
908 if ((offset > len) || (len - offset < size))
909 return -EIO;
911 return 0;
914 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
915 int nb_sectors)
917 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
918 nb_sectors * BDRV_SECTOR_SIZE);
921 /* return < 0 if error. See bdrv_write() for the return codes */
922 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
923 uint8_t *buf, int nb_sectors)
925 BlockDriver *drv = bs->drv;
927 if (!drv)
928 return -ENOMEDIUM;
929 if (bdrv_check_request(bs, sector_num, nb_sectors))
930 return -EIO;
932 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
935 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
936 int nb_sectors, int dirty)
938 int64_t start, end;
939 unsigned long val, idx, bit;
941 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
942 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
944 for (; start <= end; start++) {
945 idx = start / (sizeof(unsigned long) * 8);
946 bit = start % (sizeof(unsigned long) * 8);
947 val = bs->dirty_bitmap[idx];
948 if (dirty) {
949 if (!(val & (1UL << bit))) {
950 bs->dirty_count++;
951 val |= 1UL << bit;
953 } else {
954 if (val & (1UL << bit)) {
955 bs->dirty_count--;
956 val &= ~(1UL << bit);
959 bs->dirty_bitmap[idx] = val;
963 /* Return < 0 if error. Important errors are:
964 -EIO generic I/O error (may happen for all errors)
965 -ENOMEDIUM No media inserted.
966 -EINVAL Invalid sector number or nb_sectors
967 -EACCES Trying to write a read-only device
969 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
970 const uint8_t *buf, int nb_sectors)
972 BlockDriver *drv = bs->drv;
973 if (!bs->drv)
974 return -ENOMEDIUM;
975 if (bs->read_only)
976 return -EACCES;
977 if (bdrv_check_request(bs, sector_num, nb_sectors))
978 return -EIO;
980 if (bs->dirty_bitmap) {
981 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
984 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
985 bs->wr_highest_sector = sector_num + nb_sectors - 1;
988 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
991 int bdrv_pread(BlockDriverState *bs, int64_t offset,
992 void *buf, int count1)
994 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
995 int len, nb_sectors, count;
996 int64_t sector_num;
997 int ret;
999 count = count1;
1000 /* first read to align to sector start */
1001 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1002 if (len > count)
1003 len = count;
1004 sector_num = offset >> BDRV_SECTOR_BITS;
1005 if (len > 0) {
1006 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1007 return ret;
1008 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1009 count -= len;
1010 if (count == 0)
1011 return count1;
1012 sector_num++;
1013 buf += len;
1016 /* read the sectors "in place" */
1017 nb_sectors = count >> BDRV_SECTOR_BITS;
1018 if (nb_sectors > 0) {
1019 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1020 return ret;
1021 sector_num += nb_sectors;
1022 len = nb_sectors << BDRV_SECTOR_BITS;
1023 buf += len;
1024 count -= len;
1027 /* add data from the last sector */
1028 if (count > 0) {
1029 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1030 return ret;
1031 memcpy(buf, tmp_buf, count);
1033 return count1;
1036 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1037 const void *buf, int count1)
1039 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1040 int len, nb_sectors, count;
1041 int64_t sector_num;
1042 int ret;
1044 count = count1;
1045 /* first write to align to sector start */
1046 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1047 if (len > count)
1048 len = count;
1049 sector_num = offset >> BDRV_SECTOR_BITS;
1050 if (len > 0) {
1051 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1052 return ret;
1053 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1054 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1055 return ret;
1056 count -= len;
1057 if (count == 0)
1058 return count1;
1059 sector_num++;
1060 buf += len;
1063 /* write the sectors "in place" */
1064 nb_sectors = count >> BDRV_SECTOR_BITS;
1065 if (nb_sectors > 0) {
1066 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1067 return ret;
1068 sector_num += nb_sectors;
1069 len = nb_sectors << BDRV_SECTOR_BITS;
1070 buf += len;
1071 count -= len;
1074 /* add data from the last sector */
1075 if (count > 0) {
1076 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1077 return ret;
1078 memcpy(tmp_buf, buf, count);
1079 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1080 return ret;
1082 return count1;
1086 * Writes to the file and ensures that no writes are reordered across this
1087 * request (acts as a barrier)
1089 * Returns 0 on success, -errno in error cases.
1091 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1092 const void *buf, int count)
1094 int ret;
1096 ret = bdrv_pwrite(bs, offset, buf, count);
1097 if (ret < 0) {
1098 return ret;
1101 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1102 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1103 bdrv_flush(bs);
1106 return 0;
1110 * Writes to the file and ensures that no writes are reordered across this
1111 * request (acts as a barrier)
1113 * Returns 0 on success, -errno in error cases.
1115 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1116 const uint8_t *buf, int nb_sectors)
1118 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1119 buf, BDRV_SECTOR_SIZE * nb_sectors);
1123 * Truncate file to 'offset' bytes (needed only for file protocols)
1125 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1127 BlockDriver *drv = bs->drv;
1128 int ret;
1129 if (!drv)
1130 return -ENOMEDIUM;
1131 if (!drv->bdrv_truncate)
1132 return -ENOTSUP;
1133 if (bs->read_only)
1134 return -EACCES;
1135 if (bdrv_in_use(bs))
1136 return -EBUSY;
1137 ret = drv->bdrv_truncate(bs, offset);
1138 if (ret == 0) {
1139 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1140 if (bs->change_cb) {
1141 bs->change_cb(bs->change_opaque, CHANGE_SIZE);
1144 return ret;
1148 * Length of a file in bytes. Return < 0 if error or unknown.
1150 int64_t bdrv_getlength(BlockDriverState *bs)
1152 BlockDriver *drv = bs->drv;
1153 if (!drv)
1154 return -ENOMEDIUM;
1156 /* Fixed size devices use the total_sectors value for speed instead of
1157 issuing a length query (like lseek) on each call. Also, legacy block
1158 drivers don't provide a bdrv_getlength function and must use
1159 total_sectors. */
1160 if (!bs->growable || !drv->bdrv_getlength) {
1161 return bs->total_sectors * BDRV_SECTOR_SIZE;
1163 return drv->bdrv_getlength(bs);
1166 /* return 0 as number of sectors if no device present or error */
1167 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1169 int64_t length;
1170 length = bdrv_getlength(bs);
1171 if (length < 0)
1172 length = 0;
1173 else
1174 length = length >> BDRV_SECTOR_BITS;
1175 *nb_sectors_ptr = length;
1178 struct partition {
1179 uint8_t boot_ind; /* 0x80 - active */
1180 uint8_t head; /* starting head */
1181 uint8_t sector; /* starting sector */
1182 uint8_t cyl; /* starting cylinder */
1183 uint8_t sys_ind; /* What partition type */
1184 uint8_t end_head; /* end head */
1185 uint8_t end_sector; /* end sector */
1186 uint8_t end_cyl; /* end cylinder */
1187 uint32_t start_sect; /* starting sector counting from 0 */
1188 uint32_t nr_sects; /* nr of sectors in partition */
1189 } __attribute__((packed));
1191 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1192 static int guess_disk_lchs(BlockDriverState *bs,
1193 int *pcylinders, int *pheads, int *psectors)
1195 uint8_t buf[BDRV_SECTOR_SIZE];
1196 int ret, i, heads, sectors, cylinders;
1197 struct partition *p;
1198 uint32_t nr_sects;
1199 uint64_t nb_sectors;
1201 bdrv_get_geometry(bs, &nb_sectors);
1203 ret = bdrv_read(bs, 0, buf, 1);
1204 if (ret < 0)
1205 return -1;
1206 /* test msdos magic */
1207 if (buf[510] != 0x55 || buf[511] != 0xaa)
1208 return -1;
1209 for(i = 0; i < 4; i++) {
1210 p = ((struct partition *)(buf + 0x1be)) + i;
1211 nr_sects = le32_to_cpu(p->nr_sects);
1212 if (nr_sects && p->end_head) {
1213 /* We make the assumption that the partition terminates on
1214 a cylinder boundary */
1215 heads = p->end_head + 1;
1216 sectors = p->end_sector & 63;
1217 if (sectors == 0)
1218 continue;
1219 cylinders = nb_sectors / (heads * sectors);
1220 if (cylinders < 1 || cylinders > 16383)
1221 continue;
1222 *pheads = heads;
1223 *psectors = sectors;
1224 *pcylinders = cylinders;
1225 #if 0
1226 printf("guessed geometry: LCHS=%d %d %d\n",
1227 cylinders, heads, sectors);
1228 #endif
1229 return 0;
1232 return -1;
1235 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1237 int translation, lba_detected = 0;
1238 int cylinders, heads, secs;
1239 uint64_t nb_sectors;
1241 /* if a geometry hint is available, use it */
1242 bdrv_get_geometry(bs, &nb_sectors);
1243 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1244 translation = bdrv_get_translation_hint(bs);
1245 if (cylinders != 0) {
1246 *pcyls = cylinders;
1247 *pheads = heads;
1248 *psecs = secs;
1249 } else {
1250 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1251 if (heads > 16) {
1252 /* if heads > 16, it means that a BIOS LBA
1253 translation was active, so the default
1254 hardware geometry is OK */
1255 lba_detected = 1;
1256 goto default_geometry;
1257 } else {
1258 *pcyls = cylinders;
1259 *pheads = heads;
1260 *psecs = secs;
1261 /* disable any translation to be in sync with
1262 the logical geometry */
1263 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1264 bdrv_set_translation_hint(bs,
1265 BIOS_ATA_TRANSLATION_NONE);
1268 } else {
1269 default_geometry:
1270 /* if no geometry, use a standard physical disk geometry */
1271 cylinders = nb_sectors / (16 * 63);
1273 if (cylinders > 16383)
1274 cylinders = 16383;
1275 else if (cylinders < 2)
1276 cylinders = 2;
1277 *pcyls = cylinders;
1278 *pheads = 16;
1279 *psecs = 63;
1280 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1281 if ((*pcyls * *pheads) <= 131072) {
1282 bdrv_set_translation_hint(bs,
1283 BIOS_ATA_TRANSLATION_LARGE);
1284 } else {
1285 bdrv_set_translation_hint(bs,
1286 BIOS_ATA_TRANSLATION_LBA);
1290 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1294 void bdrv_set_geometry_hint(BlockDriverState *bs,
1295 int cyls, int heads, int secs)
1297 bs->cyls = cyls;
1298 bs->heads = heads;
1299 bs->secs = secs;
1302 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1304 bs->type = type;
1305 bs->removable = ((type == BDRV_TYPE_CDROM ||
1306 type == BDRV_TYPE_FLOPPY));
1309 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1311 bs->translation = translation;
1314 void bdrv_get_geometry_hint(BlockDriverState *bs,
1315 int *pcyls, int *pheads, int *psecs)
1317 *pcyls = bs->cyls;
1318 *pheads = bs->heads;
1319 *psecs = bs->secs;
1322 int bdrv_get_type_hint(BlockDriverState *bs)
1324 return bs->type;
1327 int bdrv_get_translation_hint(BlockDriverState *bs)
1329 return bs->translation;
1332 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1333 BlockErrorAction on_write_error)
1335 bs->on_read_error = on_read_error;
1336 bs->on_write_error = on_write_error;
1339 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1341 return is_read ? bs->on_read_error : bs->on_write_error;
1344 void bdrv_set_removable(BlockDriverState *bs, int removable)
1346 bs->removable = removable;
1347 if (removable && bs == bs_snapshots) {
1348 bs_snapshots = NULL;
1352 int bdrv_is_removable(BlockDriverState *bs)
1354 return bs->removable;
1357 int bdrv_is_read_only(BlockDriverState *bs)
1359 return bs->read_only;
1362 int bdrv_is_sg(BlockDriverState *bs)
1364 return bs->sg;
1367 int bdrv_enable_write_cache(BlockDriverState *bs)
1369 return bs->enable_write_cache;
1372 /* XXX: no longer used */
1373 void bdrv_set_change_cb(BlockDriverState *bs,
1374 void (*change_cb)(void *opaque, int reason),
1375 void *opaque)
1377 bs->change_cb = change_cb;
1378 bs->change_opaque = opaque;
1381 int bdrv_is_encrypted(BlockDriverState *bs)
1383 if (bs->backing_hd && bs->backing_hd->encrypted)
1384 return 1;
1385 return bs->encrypted;
1388 int bdrv_key_required(BlockDriverState *bs)
1390 BlockDriverState *backing_hd = bs->backing_hd;
1392 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1393 return 1;
1394 return (bs->encrypted && !bs->valid_key);
1397 int bdrv_set_key(BlockDriverState *bs, const char *key)
1399 int ret;
1400 if (bs->backing_hd && bs->backing_hd->encrypted) {
1401 ret = bdrv_set_key(bs->backing_hd, key);
1402 if (ret < 0)
1403 return ret;
1404 if (!bs->encrypted)
1405 return 0;
1407 if (!bs->encrypted) {
1408 return -EINVAL;
1409 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1410 return -ENOMEDIUM;
1412 ret = bs->drv->bdrv_set_key(bs, key);
1413 if (ret < 0) {
1414 bs->valid_key = 0;
1415 } else if (!bs->valid_key) {
1416 bs->valid_key = 1;
1417 /* call the change callback now, we skipped it on open */
1418 bs->media_changed = 1;
1419 if (bs->change_cb)
1420 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
1422 return ret;
1425 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1427 if (!bs->drv) {
1428 buf[0] = '\0';
1429 } else {
1430 pstrcpy(buf, buf_size, bs->drv->format_name);
1434 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1435 void *opaque)
1437 BlockDriver *drv;
1439 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1440 it(opaque, drv->format_name);
1444 BlockDriverState *bdrv_find(const char *name)
1446 BlockDriverState *bs;
1448 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1449 if (!strcmp(name, bs->device_name)) {
1450 return bs;
1453 return NULL;
1456 BlockDriverState *bdrv_next(BlockDriverState *bs)
1458 if (!bs) {
1459 return QTAILQ_FIRST(&bdrv_states);
1461 return QTAILQ_NEXT(bs, list);
1464 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1466 BlockDriverState *bs;
1468 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1469 it(opaque, bs);
1473 const char *bdrv_get_device_name(BlockDriverState *bs)
1475 return bs->device_name;
1478 int bdrv_flush(BlockDriverState *bs)
1480 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1481 return 0;
1484 if (bs->drv && bs->drv->bdrv_flush) {
1485 return bs->drv->bdrv_flush(bs);
1489 * Some block drivers always operate in either writethrough or unsafe mode
1490 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1491 * the server works (because the behaviour is hardcoded or depends on
1492 * server-side configuration), so we can't ensure that everything is safe
1493 * on disk. Returning an error doesn't work because that would break guests
1494 * even if the server operates in writethrough mode.
1496 * Let's hope the user knows what he's doing.
1498 return 0;
1501 void bdrv_flush_all(void)
1503 BlockDriverState *bs;
1505 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1506 if (bs->drv && !bdrv_is_read_only(bs) &&
1507 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1508 bdrv_flush(bs);
1513 int bdrv_has_zero_init(BlockDriverState *bs)
1515 assert(bs->drv);
1517 if (bs->drv->bdrv_has_zero_init) {
1518 return bs->drv->bdrv_has_zero_init(bs);
1521 return 1;
1524 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1526 if (!bs->drv) {
1527 return -ENOMEDIUM;
1529 if (!bs->drv->bdrv_discard) {
1530 return 0;
1532 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1536 * Returns true iff the specified sector is present in the disk image. Drivers
1537 * not implementing the functionality are assumed to not support backing files,
1538 * hence all their sectors are reported as allocated.
1540 * 'pnum' is set to the number of sectors (including and immediately following
1541 * the specified sector) that are known to be in the same
1542 * allocated/unallocated state.
1544 * 'nb_sectors' is the max value 'pnum' should be set to.
1546 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1547 int *pnum)
1549 int64_t n;
1550 if (!bs->drv->bdrv_is_allocated) {
1551 if (sector_num >= bs->total_sectors) {
1552 *pnum = 0;
1553 return 0;
1555 n = bs->total_sectors - sector_num;
1556 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1557 return 1;
1559 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1562 void bdrv_mon_event(const BlockDriverState *bdrv,
1563 BlockMonEventAction action, int is_read)
1565 QObject *data;
1566 const char *action_str;
1568 switch (action) {
1569 case BDRV_ACTION_REPORT:
1570 action_str = "report";
1571 break;
1572 case BDRV_ACTION_IGNORE:
1573 action_str = "ignore";
1574 break;
1575 case BDRV_ACTION_STOP:
1576 action_str = "stop";
1577 break;
1578 default:
1579 abort();
1582 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1583 bdrv->device_name,
1584 action_str,
1585 is_read ? "read" : "write");
1586 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1588 qobject_decref(data);
1591 static void bdrv_print_dict(QObject *obj, void *opaque)
1593 QDict *bs_dict;
1594 Monitor *mon = opaque;
1596 bs_dict = qobject_to_qdict(obj);
1598 monitor_printf(mon, "%s: type=%s removable=%d",
1599 qdict_get_str(bs_dict, "device"),
1600 qdict_get_str(bs_dict, "type"),
1601 qdict_get_bool(bs_dict, "removable"));
1603 if (qdict_get_bool(bs_dict, "removable")) {
1604 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1607 if (qdict_haskey(bs_dict, "inserted")) {
1608 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1610 monitor_printf(mon, " file=");
1611 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1612 if (qdict_haskey(qdict, "backing_file")) {
1613 monitor_printf(mon, " backing_file=");
1614 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1616 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1617 qdict_get_bool(qdict, "ro"),
1618 qdict_get_str(qdict, "drv"),
1619 qdict_get_bool(qdict, "encrypted"));
1620 } else {
1621 monitor_printf(mon, " [not inserted]");
1624 monitor_printf(mon, "\n");
1627 void bdrv_info_print(Monitor *mon, const QObject *data)
1629 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1632 void bdrv_info(Monitor *mon, QObject **ret_data)
1634 QList *bs_list;
1635 BlockDriverState *bs;
1637 bs_list = qlist_new();
1639 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1640 QObject *bs_obj;
1641 const char *type = "unknown";
1643 switch(bs->type) {
1644 case BDRV_TYPE_HD:
1645 type = "hd";
1646 break;
1647 case BDRV_TYPE_CDROM:
1648 type = "cdrom";
1649 break;
1650 case BDRV_TYPE_FLOPPY:
1651 type = "floppy";
1652 break;
1655 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1656 "'removable': %i, 'locked': %i }",
1657 bs->device_name, type, bs->removable,
1658 bs->locked);
1660 if (bs->drv) {
1661 QObject *obj;
1662 QDict *bs_dict = qobject_to_qdict(bs_obj);
1664 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1665 "'encrypted': %i }",
1666 bs->filename, bs->read_only,
1667 bs->drv->format_name,
1668 bdrv_is_encrypted(bs));
1669 if (bs->backing_file[0] != '\0') {
1670 QDict *qdict = qobject_to_qdict(obj);
1671 qdict_put(qdict, "backing_file",
1672 qstring_from_str(bs->backing_file));
1675 qdict_put_obj(bs_dict, "inserted", obj);
1677 qlist_append_obj(bs_list, bs_obj);
1680 *ret_data = QOBJECT(bs_list);
1683 static void bdrv_stats_iter(QObject *data, void *opaque)
1685 QDict *qdict;
1686 Monitor *mon = opaque;
1688 qdict = qobject_to_qdict(data);
1689 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1691 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1692 monitor_printf(mon, " rd_bytes=%" PRId64
1693 " wr_bytes=%" PRId64
1694 " rd_operations=%" PRId64
1695 " wr_operations=%" PRId64
1696 "\n",
1697 qdict_get_int(qdict, "rd_bytes"),
1698 qdict_get_int(qdict, "wr_bytes"),
1699 qdict_get_int(qdict, "rd_operations"),
1700 qdict_get_int(qdict, "wr_operations"));
1703 void bdrv_stats_print(Monitor *mon, const QObject *data)
1705 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1708 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1710 QObject *res;
1711 QDict *dict;
1713 res = qobject_from_jsonf("{ 'stats': {"
1714 "'rd_bytes': %" PRId64 ","
1715 "'wr_bytes': %" PRId64 ","
1716 "'rd_operations': %" PRId64 ","
1717 "'wr_operations': %" PRId64 ","
1718 "'wr_highest_offset': %" PRId64
1719 "} }",
1720 bs->rd_bytes, bs->wr_bytes,
1721 bs->rd_ops, bs->wr_ops,
1722 bs->wr_highest_sector *
1723 (uint64_t)BDRV_SECTOR_SIZE);
1724 dict = qobject_to_qdict(res);
1726 if (*bs->device_name) {
1727 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1730 if (bs->file) {
1731 QObject *parent = bdrv_info_stats_bs(bs->file);
1732 qdict_put_obj(dict, "parent", parent);
1735 return res;
1738 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1740 QObject *obj;
1741 QList *devices;
1742 BlockDriverState *bs;
1744 devices = qlist_new();
1746 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1747 obj = bdrv_info_stats_bs(bs);
1748 qlist_append_obj(devices, obj);
1751 *ret_data = QOBJECT(devices);
1754 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1756 if (bs->backing_hd && bs->backing_hd->encrypted)
1757 return bs->backing_file;
1758 else if (bs->encrypted)
1759 return bs->filename;
1760 else
1761 return NULL;
1764 void bdrv_get_backing_filename(BlockDriverState *bs,
1765 char *filename, int filename_size)
1767 if (!bs->backing_file) {
1768 pstrcpy(filename, filename_size, "");
1769 } else {
1770 pstrcpy(filename, filename_size, bs->backing_file);
1774 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1775 const uint8_t *buf, int nb_sectors)
1777 BlockDriver *drv = bs->drv;
1778 if (!drv)
1779 return -ENOMEDIUM;
1780 if (!drv->bdrv_write_compressed)
1781 return -ENOTSUP;
1782 if (bdrv_check_request(bs, sector_num, nb_sectors))
1783 return -EIO;
1785 if (bs->dirty_bitmap) {
1786 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1789 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1792 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1794 BlockDriver *drv = bs->drv;
1795 if (!drv)
1796 return -ENOMEDIUM;
1797 if (!drv->bdrv_get_info)
1798 return -ENOTSUP;
1799 memset(bdi, 0, sizeof(*bdi));
1800 return drv->bdrv_get_info(bs, bdi);
1803 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1804 int64_t pos, int size)
1806 BlockDriver *drv = bs->drv;
1807 if (!drv)
1808 return -ENOMEDIUM;
1809 if (drv->bdrv_save_vmstate)
1810 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1811 if (bs->file)
1812 return bdrv_save_vmstate(bs->file, buf, pos, size);
1813 return -ENOTSUP;
1816 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1817 int64_t pos, int size)
1819 BlockDriver *drv = bs->drv;
1820 if (!drv)
1821 return -ENOMEDIUM;
1822 if (drv->bdrv_load_vmstate)
1823 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1824 if (bs->file)
1825 return bdrv_load_vmstate(bs->file, buf, pos, size);
1826 return -ENOTSUP;
1829 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1831 BlockDriver *drv = bs->drv;
1833 if (!drv || !drv->bdrv_debug_event) {
1834 return;
1837 return drv->bdrv_debug_event(bs, event);
1841 /**************************************************************/
1842 /* handling of snapshots */
1844 int bdrv_can_snapshot(BlockDriverState *bs)
1846 BlockDriver *drv = bs->drv;
1847 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1848 return 0;
1851 if (!drv->bdrv_snapshot_create) {
1852 if (bs->file != NULL) {
1853 return bdrv_can_snapshot(bs->file);
1855 return 0;
1858 return 1;
1861 int bdrv_is_snapshot(BlockDriverState *bs)
1863 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1866 BlockDriverState *bdrv_snapshots(void)
1868 BlockDriverState *bs;
1870 if (bs_snapshots) {
1871 return bs_snapshots;
1874 bs = NULL;
1875 while ((bs = bdrv_next(bs))) {
1876 if (bdrv_can_snapshot(bs)) {
1877 bs_snapshots = bs;
1878 return bs;
1881 return NULL;
1884 int bdrv_snapshot_create(BlockDriverState *bs,
1885 QEMUSnapshotInfo *sn_info)
1887 BlockDriver *drv = bs->drv;
1888 if (!drv)
1889 return -ENOMEDIUM;
1890 if (drv->bdrv_snapshot_create)
1891 return drv->bdrv_snapshot_create(bs, sn_info);
1892 if (bs->file)
1893 return bdrv_snapshot_create(bs->file, sn_info);
1894 return -ENOTSUP;
1897 int bdrv_snapshot_goto(BlockDriverState *bs,
1898 const char *snapshot_id)
1900 BlockDriver *drv = bs->drv;
1901 int ret, open_ret;
1903 if (!drv)
1904 return -ENOMEDIUM;
1905 if (drv->bdrv_snapshot_goto)
1906 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1908 if (bs->file) {
1909 drv->bdrv_close(bs);
1910 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1911 open_ret = drv->bdrv_open(bs, bs->open_flags);
1912 if (open_ret < 0) {
1913 bdrv_delete(bs->file);
1914 bs->drv = NULL;
1915 return open_ret;
1917 return ret;
1920 return -ENOTSUP;
1923 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1925 BlockDriver *drv = bs->drv;
1926 if (!drv)
1927 return -ENOMEDIUM;
1928 if (drv->bdrv_snapshot_delete)
1929 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1930 if (bs->file)
1931 return bdrv_snapshot_delete(bs->file, snapshot_id);
1932 return -ENOTSUP;
1935 int bdrv_snapshot_list(BlockDriverState *bs,
1936 QEMUSnapshotInfo **psn_info)
1938 BlockDriver *drv = bs->drv;
1939 if (!drv)
1940 return -ENOMEDIUM;
1941 if (drv->bdrv_snapshot_list)
1942 return drv->bdrv_snapshot_list(bs, psn_info);
1943 if (bs->file)
1944 return bdrv_snapshot_list(bs->file, psn_info);
1945 return -ENOTSUP;
1948 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
1949 const char *snapshot_name)
1951 BlockDriver *drv = bs->drv;
1952 if (!drv) {
1953 return -ENOMEDIUM;
1955 if (!bs->read_only) {
1956 return -EINVAL;
1958 if (drv->bdrv_snapshot_load_tmp) {
1959 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
1961 return -ENOTSUP;
1964 #define NB_SUFFIXES 4
1966 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1968 static const char suffixes[NB_SUFFIXES] = "KMGT";
1969 int64_t base;
1970 int i;
1972 if (size <= 999) {
1973 snprintf(buf, buf_size, "%" PRId64, size);
1974 } else {
1975 base = 1024;
1976 for(i = 0; i < NB_SUFFIXES; i++) {
1977 if (size < (10 * base)) {
1978 snprintf(buf, buf_size, "%0.1f%c",
1979 (double)size / base,
1980 suffixes[i]);
1981 break;
1982 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1983 snprintf(buf, buf_size, "%" PRId64 "%c",
1984 ((size + (base >> 1)) / base),
1985 suffixes[i]);
1986 break;
1988 base = base * 1024;
1991 return buf;
1994 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1996 char buf1[128], date_buf[128], clock_buf[128];
1997 #ifdef _WIN32
1998 struct tm *ptm;
1999 #else
2000 struct tm tm;
2001 #endif
2002 time_t ti;
2003 int64_t secs;
2005 if (!sn) {
2006 snprintf(buf, buf_size,
2007 "%-10s%-20s%7s%20s%15s",
2008 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2009 } else {
2010 ti = sn->date_sec;
2011 #ifdef _WIN32
2012 ptm = localtime(&ti);
2013 strftime(date_buf, sizeof(date_buf),
2014 "%Y-%m-%d %H:%M:%S", ptm);
2015 #else
2016 localtime_r(&ti, &tm);
2017 strftime(date_buf, sizeof(date_buf),
2018 "%Y-%m-%d %H:%M:%S", &tm);
2019 #endif
2020 secs = sn->vm_clock_nsec / 1000000000;
2021 snprintf(clock_buf, sizeof(clock_buf),
2022 "%02d:%02d:%02d.%03d",
2023 (int)(secs / 3600),
2024 (int)((secs / 60) % 60),
2025 (int)(secs % 60),
2026 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2027 snprintf(buf, buf_size,
2028 "%-10s%-20s%7s%20s%15s",
2029 sn->id_str, sn->name,
2030 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2031 date_buf,
2032 clock_buf);
2034 return buf;
2038 /**************************************************************/
2039 /* async I/Os */
2041 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2042 QEMUIOVector *qiov, int nb_sectors,
2043 BlockDriverCompletionFunc *cb, void *opaque)
2045 BlockDriver *drv = bs->drv;
2046 BlockDriverAIOCB *ret;
2048 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2050 if (!drv)
2051 return NULL;
2052 if (bdrv_check_request(bs, sector_num, nb_sectors))
2053 return NULL;
2055 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2056 cb, opaque);
2058 if (ret) {
2059 /* Update stats even though technically transfer has not happened. */
2060 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2061 bs->rd_ops ++;
2064 return ret;
2067 typedef struct BlockCompleteData {
2068 BlockDriverCompletionFunc *cb;
2069 void *opaque;
2070 BlockDriverState *bs;
2071 int64_t sector_num;
2072 int nb_sectors;
2073 } BlockCompleteData;
2075 static void block_complete_cb(void *opaque, int ret)
2077 BlockCompleteData *b = opaque;
2079 if (b->bs->dirty_bitmap) {
2080 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2082 b->cb(b->opaque, ret);
2083 qemu_free(b);
2086 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2087 int64_t sector_num,
2088 int nb_sectors,
2089 BlockDriverCompletionFunc *cb,
2090 void *opaque)
2092 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2094 blkdata->bs = bs;
2095 blkdata->cb = cb;
2096 blkdata->opaque = opaque;
2097 blkdata->sector_num = sector_num;
2098 blkdata->nb_sectors = nb_sectors;
2100 return blkdata;
2103 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2104 QEMUIOVector *qiov, int nb_sectors,
2105 BlockDriverCompletionFunc *cb, void *opaque)
2107 BlockDriver *drv = bs->drv;
2108 BlockDriverAIOCB *ret;
2109 BlockCompleteData *blk_cb_data;
2111 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2113 if (!drv)
2114 return NULL;
2115 if (bs->read_only)
2116 return NULL;
2117 if (bdrv_check_request(bs, sector_num, nb_sectors))
2118 return NULL;
2120 if (bs->dirty_bitmap) {
2121 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2122 opaque);
2123 cb = &block_complete_cb;
2124 opaque = blk_cb_data;
2127 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2128 cb, opaque);
2130 if (ret) {
2131 /* Update stats even though technically transfer has not happened. */
2132 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2133 bs->wr_ops ++;
2134 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2135 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2139 return ret;
2143 typedef struct MultiwriteCB {
2144 int error;
2145 int num_requests;
2146 int num_callbacks;
2147 struct {
2148 BlockDriverCompletionFunc *cb;
2149 void *opaque;
2150 QEMUIOVector *free_qiov;
2151 void *free_buf;
2152 } callbacks[];
2153 } MultiwriteCB;
2155 static void multiwrite_user_cb(MultiwriteCB *mcb)
2157 int i;
2159 for (i = 0; i < mcb->num_callbacks; i++) {
2160 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2161 if (mcb->callbacks[i].free_qiov) {
2162 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2164 qemu_free(mcb->callbacks[i].free_qiov);
2165 qemu_vfree(mcb->callbacks[i].free_buf);
2169 static void multiwrite_cb(void *opaque, int ret)
2171 MultiwriteCB *mcb = opaque;
2173 trace_multiwrite_cb(mcb, ret);
2175 if (ret < 0 && !mcb->error) {
2176 mcb->error = ret;
2179 mcb->num_requests--;
2180 if (mcb->num_requests == 0) {
2181 multiwrite_user_cb(mcb);
2182 qemu_free(mcb);
2186 static int multiwrite_req_compare(const void *a, const void *b)
2188 const BlockRequest *req1 = a, *req2 = b;
2191 * Note that we can't simply subtract req2->sector from req1->sector
2192 * here as that could overflow the return value.
2194 if (req1->sector > req2->sector) {
2195 return 1;
2196 } else if (req1->sector < req2->sector) {
2197 return -1;
2198 } else {
2199 return 0;
2204 * Takes a bunch of requests and tries to merge them. Returns the number of
2205 * requests that remain after merging.
2207 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2208 int num_reqs, MultiwriteCB *mcb)
2210 int i, outidx;
2212 // Sort requests by start sector
2213 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2215 // Check if adjacent requests touch the same clusters. If so, combine them,
2216 // filling up gaps with zero sectors.
2217 outidx = 0;
2218 for (i = 1; i < num_reqs; i++) {
2219 int merge = 0;
2220 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2222 // This handles the cases that are valid for all block drivers, namely
2223 // exactly sequential writes and overlapping writes.
2224 if (reqs[i].sector <= oldreq_last) {
2225 merge = 1;
2228 // The block driver may decide that it makes sense to combine requests
2229 // even if there is a gap of some sectors between them. In this case,
2230 // the gap is filled with zeros (therefore only applicable for yet
2231 // unused space in format like qcow2).
2232 if (!merge && bs->drv->bdrv_merge_requests) {
2233 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2236 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2237 merge = 0;
2240 if (merge) {
2241 size_t size;
2242 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2243 qemu_iovec_init(qiov,
2244 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2246 // Add the first request to the merged one. If the requests are
2247 // overlapping, drop the last sectors of the first request.
2248 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2249 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2251 // We might need to add some zeros between the two requests
2252 if (reqs[i].sector > oldreq_last) {
2253 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2254 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2255 memset(buf, 0, zero_bytes);
2256 qemu_iovec_add(qiov, buf, zero_bytes);
2257 mcb->callbacks[i].free_buf = buf;
2260 // Add the second request
2261 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2263 reqs[outidx].nb_sectors = qiov->size >> 9;
2264 reqs[outidx].qiov = qiov;
2266 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2267 } else {
2268 outidx++;
2269 reqs[outidx].sector = reqs[i].sector;
2270 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2271 reqs[outidx].qiov = reqs[i].qiov;
2275 return outidx + 1;
2279 * Submit multiple AIO write requests at once.
2281 * On success, the function returns 0 and all requests in the reqs array have
2282 * been submitted. In error case this function returns -1, and any of the
2283 * requests may or may not be submitted yet. In particular, this means that the
2284 * callback will be called for some of the requests, for others it won't. The
2285 * caller must check the error field of the BlockRequest to wait for the right
2286 * callbacks (if error != 0, no callback will be called).
2288 * The implementation may modify the contents of the reqs array, e.g. to merge
2289 * requests. However, the fields opaque and error are left unmodified as they
2290 * are used to signal failure for a single request to the caller.
2292 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2294 BlockDriverAIOCB *acb;
2295 MultiwriteCB *mcb;
2296 int i;
2298 if (num_reqs == 0) {
2299 return 0;
2302 // Create MultiwriteCB structure
2303 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2304 mcb->num_requests = 0;
2305 mcb->num_callbacks = num_reqs;
2307 for (i = 0; i < num_reqs; i++) {
2308 mcb->callbacks[i].cb = reqs[i].cb;
2309 mcb->callbacks[i].opaque = reqs[i].opaque;
2312 // Check for mergable requests
2313 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2315 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2318 * Run the aio requests. As soon as one request can't be submitted
2319 * successfully, fail all requests that are not yet submitted (we must
2320 * return failure for all requests anyway)
2322 * num_requests cannot be set to the right value immediately: If
2323 * bdrv_aio_writev fails for some request, num_requests would be too high
2324 * and therefore multiwrite_cb() would never recognize the multiwrite
2325 * request as completed. We also cannot use the loop variable i to set it
2326 * when the first request fails because the callback may already have been
2327 * called for previously submitted requests. Thus, num_requests must be
2328 * incremented for each request that is submitted.
2330 * The problem that callbacks may be called early also means that we need
2331 * to take care that num_requests doesn't become 0 before all requests are
2332 * submitted - multiwrite_cb() would consider the multiwrite request
2333 * completed. A dummy request that is "completed" by a manual call to
2334 * multiwrite_cb() takes care of this.
2336 mcb->num_requests = 1;
2338 // Run the aio requests
2339 for (i = 0; i < num_reqs; i++) {
2340 mcb->num_requests++;
2341 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2342 reqs[i].nb_sectors, multiwrite_cb, mcb);
2344 if (acb == NULL) {
2345 // We can only fail the whole thing if no request has been
2346 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2347 // complete and report the error in the callback.
2348 if (i == 0) {
2349 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2350 goto fail;
2351 } else {
2352 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2353 multiwrite_cb(mcb, -EIO);
2354 break;
2359 /* Complete the dummy request */
2360 multiwrite_cb(mcb, 0);
2362 return 0;
2364 fail:
2365 for (i = 0; i < mcb->num_callbacks; i++) {
2366 reqs[i].error = -EIO;
2368 qemu_free(mcb);
2369 return -1;
2372 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2373 BlockDriverCompletionFunc *cb, void *opaque)
2375 BlockDriver *drv = bs->drv;
2377 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2378 return bdrv_aio_noop_em(bs, cb, opaque);
2381 if (!drv)
2382 return NULL;
2383 return drv->bdrv_aio_flush(bs, cb, opaque);
2386 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2388 acb->pool->cancel(acb);
2392 /**************************************************************/
2393 /* async block device emulation */
2395 typedef struct BlockDriverAIOCBSync {
2396 BlockDriverAIOCB common;
2397 QEMUBH *bh;
2398 int ret;
2399 /* vector translation state */
2400 QEMUIOVector *qiov;
2401 uint8_t *bounce;
2402 int is_write;
2403 } BlockDriverAIOCBSync;
2405 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2407 BlockDriverAIOCBSync *acb =
2408 container_of(blockacb, BlockDriverAIOCBSync, common);
2409 qemu_bh_delete(acb->bh);
2410 acb->bh = NULL;
2411 qemu_aio_release(acb);
2414 static AIOPool bdrv_em_aio_pool = {
2415 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2416 .cancel = bdrv_aio_cancel_em,
2419 static void bdrv_aio_bh_cb(void *opaque)
2421 BlockDriverAIOCBSync *acb = opaque;
2423 if (!acb->is_write)
2424 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2425 qemu_vfree(acb->bounce);
2426 acb->common.cb(acb->common.opaque, acb->ret);
2427 qemu_bh_delete(acb->bh);
2428 acb->bh = NULL;
2429 qemu_aio_release(acb);
2432 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2433 int64_t sector_num,
2434 QEMUIOVector *qiov,
2435 int nb_sectors,
2436 BlockDriverCompletionFunc *cb,
2437 void *opaque,
2438 int is_write)
2441 BlockDriverAIOCBSync *acb;
2443 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2444 acb->is_write = is_write;
2445 acb->qiov = qiov;
2446 acb->bounce = qemu_blockalign(bs, qiov->size);
2448 if (!acb->bh)
2449 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2451 if (is_write) {
2452 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2453 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2454 } else {
2455 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2458 qemu_bh_schedule(acb->bh);
2460 return &acb->common;
2463 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2464 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2465 BlockDriverCompletionFunc *cb, void *opaque)
2467 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2470 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2471 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2472 BlockDriverCompletionFunc *cb, void *opaque)
2474 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2477 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2478 BlockDriverCompletionFunc *cb, void *opaque)
2480 BlockDriverAIOCBSync *acb;
2482 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2483 acb->is_write = 1; /* don't bounce in the completion hadler */
2484 acb->qiov = NULL;
2485 acb->bounce = NULL;
2486 acb->ret = 0;
2488 if (!acb->bh)
2489 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2491 bdrv_flush(bs);
2492 qemu_bh_schedule(acb->bh);
2493 return &acb->common;
2496 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2497 BlockDriverCompletionFunc *cb, void *opaque)
2499 BlockDriverAIOCBSync *acb;
2501 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2502 acb->is_write = 1; /* don't bounce in the completion handler */
2503 acb->qiov = NULL;
2504 acb->bounce = NULL;
2505 acb->ret = 0;
2507 if (!acb->bh) {
2508 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2511 qemu_bh_schedule(acb->bh);
2512 return &acb->common;
2515 /**************************************************************/
2516 /* sync block device emulation */
2518 static void bdrv_rw_em_cb(void *opaque, int ret)
2520 *(int *)opaque = ret;
2523 #define NOT_DONE 0x7fffffff
2525 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2526 uint8_t *buf, int nb_sectors)
2528 int async_ret;
2529 BlockDriverAIOCB *acb;
2530 struct iovec iov;
2531 QEMUIOVector qiov;
2533 async_context_push();
2535 async_ret = NOT_DONE;
2536 iov.iov_base = (void *)buf;
2537 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2538 qemu_iovec_init_external(&qiov, &iov, 1);
2539 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2540 bdrv_rw_em_cb, &async_ret);
2541 if (acb == NULL) {
2542 async_ret = -1;
2543 goto fail;
2546 while (async_ret == NOT_DONE) {
2547 qemu_aio_wait();
2551 fail:
2552 async_context_pop();
2553 return async_ret;
2556 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2557 const uint8_t *buf, int nb_sectors)
2559 int async_ret;
2560 BlockDriverAIOCB *acb;
2561 struct iovec iov;
2562 QEMUIOVector qiov;
2564 async_context_push();
2566 async_ret = NOT_DONE;
2567 iov.iov_base = (void *)buf;
2568 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2569 qemu_iovec_init_external(&qiov, &iov, 1);
2570 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2571 bdrv_rw_em_cb, &async_ret);
2572 if (acb == NULL) {
2573 async_ret = -1;
2574 goto fail;
2576 while (async_ret == NOT_DONE) {
2577 qemu_aio_wait();
2580 fail:
2581 async_context_pop();
2582 return async_ret;
2585 void bdrv_init(void)
2587 module_call_init(MODULE_INIT_BLOCK);
2590 void bdrv_init_with_whitelist(void)
2592 use_bdrv_whitelist = 1;
2593 bdrv_init();
2596 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2597 BlockDriverCompletionFunc *cb, void *opaque)
2599 BlockDriverAIOCB *acb;
2601 if (pool->free_aiocb) {
2602 acb = pool->free_aiocb;
2603 pool->free_aiocb = acb->next;
2604 } else {
2605 acb = qemu_mallocz(pool->aiocb_size);
2606 acb->pool = pool;
2608 acb->bs = bs;
2609 acb->cb = cb;
2610 acb->opaque = opaque;
2611 return acb;
2614 void qemu_aio_release(void *p)
2616 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2617 AIOPool *pool = acb->pool;
2618 acb->next = pool->free_aiocb;
2619 pool->free_aiocb = acb;
2622 /**************************************************************/
2623 /* removable device support */
2626 * Return TRUE if the media is present
2628 int bdrv_is_inserted(BlockDriverState *bs)
2630 BlockDriver *drv = bs->drv;
2631 int ret;
2632 if (!drv)
2633 return 0;
2634 if (!drv->bdrv_is_inserted)
2635 return !bs->tray_open;
2636 ret = drv->bdrv_is_inserted(bs);
2637 return ret;
2641 * Return TRUE if the media changed since the last call to this
2642 * function. It is currently only used for floppy disks
2644 int bdrv_media_changed(BlockDriverState *bs)
2646 BlockDriver *drv = bs->drv;
2647 int ret;
2649 if (!drv || !drv->bdrv_media_changed)
2650 ret = -ENOTSUP;
2651 else
2652 ret = drv->bdrv_media_changed(bs);
2653 if (ret == -ENOTSUP)
2654 ret = bs->media_changed;
2655 bs->media_changed = 0;
2656 return ret;
2660 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2662 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2664 BlockDriver *drv = bs->drv;
2665 int ret;
2667 if (bs->locked) {
2668 return -EBUSY;
2671 if (!drv || !drv->bdrv_eject) {
2672 ret = -ENOTSUP;
2673 } else {
2674 ret = drv->bdrv_eject(bs, eject_flag);
2676 if (ret == -ENOTSUP) {
2677 ret = 0;
2679 if (ret >= 0) {
2680 bs->tray_open = eject_flag;
2683 return ret;
2686 int bdrv_is_locked(BlockDriverState *bs)
2688 return bs->locked;
2692 * Lock or unlock the media (if it is locked, the user won't be able
2693 * to eject it manually).
2695 void bdrv_set_locked(BlockDriverState *bs, int locked)
2697 BlockDriver *drv = bs->drv;
2699 bs->locked = locked;
2700 if (drv && drv->bdrv_set_locked) {
2701 drv->bdrv_set_locked(bs, locked);
2705 /* needed for generic scsi interface */
2707 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2709 BlockDriver *drv = bs->drv;
2711 if (drv && drv->bdrv_ioctl)
2712 return drv->bdrv_ioctl(bs, req, buf);
2713 return -ENOTSUP;
2716 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2717 unsigned long int req, void *buf,
2718 BlockDriverCompletionFunc *cb, void *opaque)
2720 BlockDriver *drv = bs->drv;
2722 if (drv && drv->bdrv_aio_ioctl)
2723 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2724 return NULL;
2729 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2731 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2734 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2736 int64_t bitmap_size;
2738 bs->dirty_count = 0;
2739 if (enable) {
2740 if (!bs->dirty_bitmap) {
2741 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2742 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2743 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2745 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2747 } else {
2748 if (bs->dirty_bitmap) {
2749 qemu_free(bs->dirty_bitmap);
2750 bs->dirty_bitmap = NULL;
2755 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2757 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2759 if (bs->dirty_bitmap &&
2760 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2761 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2762 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2763 } else {
2764 return 0;
2768 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2769 int nr_sectors)
2771 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2774 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2776 return bs->dirty_count;
2779 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
2781 assert(bs->in_use != in_use);
2782 bs->in_use = in_use;
2785 int bdrv_in_use(BlockDriverState *bs)
2787 return bs->in_use;
2790 int bdrv_img_create(const char *filename, const char *fmt,
2791 const char *base_filename, const char *base_fmt,
2792 char *options, uint64_t img_size, int flags)
2794 QEMUOptionParameter *param = NULL, *create_options = NULL;
2795 QEMUOptionParameter *backing_fmt, *backing_file;
2796 BlockDriverState *bs = NULL;
2797 BlockDriver *drv, *proto_drv;
2798 BlockDriver *backing_drv = NULL;
2799 int ret = 0;
2801 /* Find driver and parse its options */
2802 drv = bdrv_find_format(fmt);
2803 if (!drv) {
2804 error_report("Unknown file format '%s'", fmt);
2805 ret = -EINVAL;
2806 goto out;
2809 proto_drv = bdrv_find_protocol(filename);
2810 if (!proto_drv) {
2811 error_report("Unknown protocol '%s'", filename);
2812 ret = -EINVAL;
2813 goto out;
2816 create_options = append_option_parameters(create_options,
2817 drv->create_options);
2818 create_options = append_option_parameters(create_options,
2819 proto_drv->create_options);
2821 /* Create parameter list with default values */
2822 param = parse_option_parameters("", create_options, param);
2824 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
2826 /* Parse -o options */
2827 if (options) {
2828 param = parse_option_parameters(options, create_options, param);
2829 if (param == NULL) {
2830 error_report("Invalid options for file format '%s'.", fmt);
2831 ret = -EINVAL;
2832 goto out;
2836 if (base_filename) {
2837 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
2838 base_filename)) {
2839 error_report("Backing file not supported for file format '%s'",
2840 fmt);
2841 ret = -EINVAL;
2842 goto out;
2846 if (base_fmt) {
2847 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
2848 error_report("Backing file format not supported for file "
2849 "format '%s'", fmt);
2850 ret = -EINVAL;
2851 goto out;
2855 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
2856 if (backing_file && backing_file->value.s) {
2857 if (!strcmp(filename, backing_file->value.s)) {
2858 error_report("Error: Trying to create an image with the "
2859 "same filename as the backing file");
2860 ret = -EINVAL;
2861 goto out;
2865 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
2866 if (backing_fmt && backing_fmt->value.s) {
2867 backing_drv = bdrv_find_format(backing_fmt->value.s);
2868 if (!backing_drv) {
2869 error_report("Unknown backing file format '%s'",
2870 backing_fmt->value.s);
2871 ret = -EINVAL;
2872 goto out;
2876 // The size for the image must always be specified, with one exception:
2877 // If we are using a backing file, we can obtain the size from there
2878 if (get_option_parameter(param, BLOCK_OPT_SIZE)->value.n == -1) {
2879 if (backing_file && backing_file->value.s) {
2880 uint64_t size;
2881 char buf[32];
2883 bs = bdrv_new("");
2885 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
2886 if (ret < 0) {
2887 error_report("Could not open '%s'", backing_file->value.s);
2888 goto out;
2890 bdrv_get_geometry(bs, &size);
2891 size *= 512;
2893 snprintf(buf, sizeof(buf), "%" PRId64, size);
2894 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
2895 } else {
2896 error_report("Image creation needs a size parameter");
2897 ret = -EINVAL;
2898 goto out;
2902 printf("Formatting '%s', fmt=%s ", filename, fmt);
2903 print_option_parameters(param);
2904 puts("");
2906 ret = bdrv_create(drv, filename, param);
2908 if (ret < 0) {
2909 if (ret == -ENOTSUP) {
2910 error_report("Formatting or formatting option not supported for "
2911 "file format '%s'", fmt);
2912 } else if (ret == -EFBIG) {
2913 error_report("The image size is too large for file format '%s'",
2914 fmt);
2915 } else {
2916 error_report("%s: error while creating %s: %s", filename, fmt,
2917 strerror(-ret));
2921 out:
2922 free_option_parameters(create_options);
2923 free_option_parameters(param);
2925 if (bs) {
2926 bdrv_delete(bs);
2929 return ret;