Correct alarm deadline computation
[qemu/ar7.git] / block.c
blob998df1b54b8c7418052e58e64fe79523cfdf9376
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 ret = drv->bdrv_truncate(bs, offset);
1136 if (ret == 0) {
1137 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1138 if (bs->change_cb) {
1139 bs->change_cb(bs->change_opaque, CHANGE_SIZE);
1142 return ret;
1146 * Length of a file in bytes. Return < 0 if error or unknown.
1148 int64_t bdrv_getlength(BlockDriverState *bs)
1150 BlockDriver *drv = bs->drv;
1151 if (!drv)
1152 return -ENOMEDIUM;
1154 /* Fixed size devices use the total_sectors value for speed instead of
1155 issuing a length query (like lseek) on each call. Also, legacy block
1156 drivers don't provide a bdrv_getlength function and must use
1157 total_sectors. */
1158 if (!bs->growable || !drv->bdrv_getlength) {
1159 return bs->total_sectors * BDRV_SECTOR_SIZE;
1161 return drv->bdrv_getlength(bs);
1164 /* return 0 as number of sectors if no device present or error */
1165 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1167 int64_t length;
1168 length = bdrv_getlength(bs);
1169 if (length < 0)
1170 length = 0;
1171 else
1172 length = length >> BDRV_SECTOR_BITS;
1173 *nb_sectors_ptr = length;
1176 struct partition {
1177 uint8_t boot_ind; /* 0x80 - active */
1178 uint8_t head; /* starting head */
1179 uint8_t sector; /* starting sector */
1180 uint8_t cyl; /* starting cylinder */
1181 uint8_t sys_ind; /* What partition type */
1182 uint8_t end_head; /* end head */
1183 uint8_t end_sector; /* end sector */
1184 uint8_t end_cyl; /* end cylinder */
1185 uint32_t start_sect; /* starting sector counting from 0 */
1186 uint32_t nr_sects; /* nr of sectors in partition */
1187 } __attribute__((packed));
1189 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1190 static int guess_disk_lchs(BlockDriverState *bs,
1191 int *pcylinders, int *pheads, int *psectors)
1193 uint8_t buf[BDRV_SECTOR_SIZE];
1194 int ret, i, heads, sectors, cylinders;
1195 struct partition *p;
1196 uint32_t nr_sects;
1197 uint64_t nb_sectors;
1199 bdrv_get_geometry(bs, &nb_sectors);
1201 ret = bdrv_read(bs, 0, buf, 1);
1202 if (ret < 0)
1203 return -1;
1204 /* test msdos magic */
1205 if (buf[510] != 0x55 || buf[511] != 0xaa)
1206 return -1;
1207 for(i = 0; i < 4; i++) {
1208 p = ((struct partition *)(buf + 0x1be)) + i;
1209 nr_sects = le32_to_cpu(p->nr_sects);
1210 if (nr_sects && p->end_head) {
1211 /* We make the assumption that the partition terminates on
1212 a cylinder boundary */
1213 heads = p->end_head + 1;
1214 sectors = p->end_sector & 63;
1215 if (sectors == 0)
1216 continue;
1217 cylinders = nb_sectors / (heads * sectors);
1218 if (cylinders < 1 || cylinders > 16383)
1219 continue;
1220 *pheads = heads;
1221 *psectors = sectors;
1222 *pcylinders = cylinders;
1223 #if 0
1224 printf("guessed geometry: LCHS=%d %d %d\n",
1225 cylinders, heads, sectors);
1226 #endif
1227 return 0;
1230 return -1;
1233 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1235 int translation, lba_detected = 0;
1236 int cylinders, heads, secs;
1237 uint64_t nb_sectors;
1239 /* if a geometry hint is available, use it */
1240 bdrv_get_geometry(bs, &nb_sectors);
1241 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1242 translation = bdrv_get_translation_hint(bs);
1243 if (cylinders != 0) {
1244 *pcyls = cylinders;
1245 *pheads = heads;
1246 *psecs = secs;
1247 } else {
1248 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1249 if (heads > 16) {
1250 /* if heads > 16, it means that a BIOS LBA
1251 translation was active, so the default
1252 hardware geometry is OK */
1253 lba_detected = 1;
1254 goto default_geometry;
1255 } else {
1256 *pcyls = cylinders;
1257 *pheads = heads;
1258 *psecs = secs;
1259 /* disable any translation to be in sync with
1260 the logical geometry */
1261 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1262 bdrv_set_translation_hint(bs,
1263 BIOS_ATA_TRANSLATION_NONE);
1266 } else {
1267 default_geometry:
1268 /* if no geometry, use a standard physical disk geometry */
1269 cylinders = nb_sectors / (16 * 63);
1271 if (cylinders > 16383)
1272 cylinders = 16383;
1273 else if (cylinders < 2)
1274 cylinders = 2;
1275 *pcyls = cylinders;
1276 *pheads = 16;
1277 *psecs = 63;
1278 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1279 if ((*pcyls * *pheads) <= 131072) {
1280 bdrv_set_translation_hint(bs,
1281 BIOS_ATA_TRANSLATION_LARGE);
1282 } else {
1283 bdrv_set_translation_hint(bs,
1284 BIOS_ATA_TRANSLATION_LBA);
1288 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1292 void bdrv_set_geometry_hint(BlockDriverState *bs,
1293 int cyls, int heads, int secs)
1295 bs->cyls = cyls;
1296 bs->heads = heads;
1297 bs->secs = secs;
1300 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1302 bs->type = type;
1303 bs->removable = ((type == BDRV_TYPE_CDROM ||
1304 type == BDRV_TYPE_FLOPPY));
1307 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1309 bs->translation = translation;
1312 void bdrv_get_geometry_hint(BlockDriverState *bs,
1313 int *pcyls, int *pheads, int *psecs)
1315 *pcyls = bs->cyls;
1316 *pheads = bs->heads;
1317 *psecs = bs->secs;
1320 int bdrv_get_type_hint(BlockDriverState *bs)
1322 return bs->type;
1325 int bdrv_get_translation_hint(BlockDriverState *bs)
1327 return bs->translation;
1330 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1331 BlockErrorAction on_write_error)
1333 bs->on_read_error = on_read_error;
1334 bs->on_write_error = on_write_error;
1337 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1339 return is_read ? bs->on_read_error : bs->on_write_error;
1342 void bdrv_set_removable(BlockDriverState *bs, int removable)
1344 bs->removable = removable;
1345 if (removable && bs == bs_snapshots) {
1346 bs_snapshots = NULL;
1350 int bdrv_is_removable(BlockDriverState *bs)
1352 return bs->removable;
1355 int bdrv_is_read_only(BlockDriverState *bs)
1357 return bs->read_only;
1360 int bdrv_is_sg(BlockDriverState *bs)
1362 return bs->sg;
1365 int bdrv_enable_write_cache(BlockDriverState *bs)
1367 return bs->enable_write_cache;
1370 /* XXX: no longer used */
1371 void bdrv_set_change_cb(BlockDriverState *bs,
1372 void (*change_cb)(void *opaque, int reason),
1373 void *opaque)
1375 bs->change_cb = change_cb;
1376 bs->change_opaque = opaque;
1379 int bdrv_is_encrypted(BlockDriverState *bs)
1381 if (bs->backing_hd && bs->backing_hd->encrypted)
1382 return 1;
1383 return bs->encrypted;
1386 int bdrv_key_required(BlockDriverState *bs)
1388 BlockDriverState *backing_hd = bs->backing_hd;
1390 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1391 return 1;
1392 return (bs->encrypted && !bs->valid_key);
1395 int bdrv_set_key(BlockDriverState *bs, const char *key)
1397 int ret;
1398 if (bs->backing_hd && bs->backing_hd->encrypted) {
1399 ret = bdrv_set_key(bs->backing_hd, key);
1400 if (ret < 0)
1401 return ret;
1402 if (!bs->encrypted)
1403 return 0;
1405 if (!bs->encrypted) {
1406 return -EINVAL;
1407 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1408 return -ENOMEDIUM;
1410 ret = bs->drv->bdrv_set_key(bs, key);
1411 if (ret < 0) {
1412 bs->valid_key = 0;
1413 } else if (!bs->valid_key) {
1414 bs->valid_key = 1;
1415 /* call the change callback now, we skipped it on open */
1416 bs->media_changed = 1;
1417 if (bs->change_cb)
1418 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
1420 return ret;
1423 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1425 if (!bs->drv) {
1426 buf[0] = '\0';
1427 } else {
1428 pstrcpy(buf, buf_size, bs->drv->format_name);
1432 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1433 void *opaque)
1435 BlockDriver *drv;
1437 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1438 it(opaque, drv->format_name);
1442 BlockDriverState *bdrv_find(const char *name)
1444 BlockDriverState *bs;
1446 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1447 if (!strcmp(name, bs->device_name)) {
1448 return bs;
1451 return NULL;
1454 BlockDriverState *bdrv_next(BlockDriverState *bs)
1456 if (!bs) {
1457 return QTAILQ_FIRST(&bdrv_states);
1459 return QTAILQ_NEXT(bs, list);
1462 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1464 BlockDriverState *bs;
1466 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1467 it(opaque, bs);
1471 const char *bdrv_get_device_name(BlockDriverState *bs)
1473 return bs->device_name;
1476 int bdrv_flush(BlockDriverState *bs)
1478 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1479 return 0;
1482 if (bs->drv && bs->drv->bdrv_flush) {
1483 return bs->drv->bdrv_flush(bs);
1487 * Some block drivers always operate in either writethrough or unsafe mode
1488 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1489 * the server works (because the behaviour is hardcoded or depends on
1490 * server-side configuration), so we can't ensure that everything is safe
1491 * on disk. Returning an error doesn't work because that would break guests
1492 * even if the server operates in writethrough mode.
1494 * Let's hope the user knows what he's doing.
1496 return 0;
1499 void bdrv_flush_all(void)
1501 BlockDriverState *bs;
1503 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1504 if (bs->drv && !bdrv_is_read_only(bs) &&
1505 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1506 bdrv_flush(bs);
1511 int bdrv_has_zero_init(BlockDriverState *bs)
1513 assert(bs->drv);
1515 if (bs->drv->bdrv_has_zero_init) {
1516 return bs->drv->bdrv_has_zero_init(bs);
1519 return 1;
1522 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1524 if (!bs->drv) {
1525 return -ENOMEDIUM;
1527 if (!bs->drv->bdrv_discard) {
1528 return 0;
1530 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1534 * Returns true iff the specified sector is present in the disk image. Drivers
1535 * not implementing the functionality are assumed to not support backing files,
1536 * hence all their sectors are reported as allocated.
1538 * 'pnum' is set to the number of sectors (including and immediately following
1539 * the specified sector) that are known to be in the same
1540 * allocated/unallocated state.
1542 * 'nb_sectors' is the max value 'pnum' should be set to.
1544 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1545 int *pnum)
1547 int64_t n;
1548 if (!bs->drv->bdrv_is_allocated) {
1549 if (sector_num >= bs->total_sectors) {
1550 *pnum = 0;
1551 return 0;
1553 n = bs->total_sectors - sector_num;
1554 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1555 return 1;
1557 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1560 void bdrv_mon_event(const BlockDriverState *bdrv,
1561 BlockMonEventAction action, int is_read)
1563 QObject *data;
1564 const char *action_str;
1566 switch (action) {
1567 case BDRV_ACTION_REPORT:
1568 action_str = "report";
1569 break;
1570 case BDRV_ACTION_IGNORE:
1571 action_str = "ignore";
1572 break;
1573 case BDRV_ACTION_STOP:
1574 action_str = "stop";
1575 break;
1576 default:
1577 abort();
1580 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1581 bdrv->device_name,
1582 action_str,
1583 is_read ? "read" : "write");
1584 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1586 qobject_decref(data);
1589 static void bdrv_print_dict(QObject *obj, void *opaque)
1591 QDict *bs_dict;
1592 Monitor *mon = opaque;
1594 bs_dict = qobject_to_qdict(obj);
1596 monitor_printf(mon, "%s: type=%s removable=%d",
1597 qdict_get_str(bs_dict, "device"),
1598 qdict_get_str(bs_dict, "type"),
1599 qdict_get_bool(bs_dict, "removable"));
1601 if (qdict_get_bool(bs_dict, "removable")) {
1602 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1605 if (qdict_haskey(bs_dict, "inserted")) {
1606 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1608 monitor_printf(mon, " file=");
1609 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1610 if (qdict_haskey(qdict, "backing_file")) {
1611 monitor_printf(mon, " backing_file=");
1612 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1614 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1615 qdict_get_bool(qdict, "ro"),
1616 qdict_get_str(qdict, "drv"),
1617 qdict_get_bool(qdict, "encrypted"));
1618 } else {
1619 monitor_printf(mon, " [not inserted]");
1622 monitor_printf(mon, "\n");
1625 void bdrv_info_print(Monitor *mon, const QObject *data)
1627 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1630 void bdrv_info(Monitor *mon, QObject **ret_data)
1632 QList *bs_list;
1633 BlockDriverState *bs;
1635 bs_list = qlist_new();
1637 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1638 QObject *bs_obj;
1639 const char *type = "unknown";
1641 switch(bs->type) {
1642 case BDRV_TYPE_HD:
1643 type = "hd";
1644 break;
1645 case BDRV_TYPE_CDROM:
1646 type = "cdrom";
1647 break;
1648 case BDRV_TYPE_FLOPPY:
1649 type = "floppy";
1650 break;
1653 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1654 "'removable': %i, 'locked': %i }",
1655 bs->device_name, type, bs->removable,
1656 bs->locked);
1658 if (bs->drv) {
1659 QObject *obj;
1660 QDict *bs_dict = qobject_to_qdict(bs_obj);
1662 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1663 "'encrypted': %i }",
1664 bs->filename, bs->read_only,
1665 bs->drv->format_name,
1666 bdrv_is_encrypted(bs));
1667 if (bs->backing_file[0] != '\0') {
1668 QDict *qdict = qobject_to_qdict(obj);
1669 qdict_put(qdict, "backing_file",
1670 qstring_from_str(bs->backing_file));
1673 qdict_put_obj(bs_dict, "inserted", obj);
1675 qlist_append_obj(bs_list, bs_obj);
1678 *ret_data = QOBJECT(bs_list);
1681 static void bdrv_stats_iter(QObject *data, void *opaque)
1683 QDict *qdict;
1684 Monitor *mon = opaque;
1686 qdict = qobject_to_qdict(data);
1687 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1689 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1690 monitor_printf(mon, " rd_bytes=%" PRId64
1691 " wr_bytes=%" PRId64
1692 " rd_operations=%" PRId64
1693 " wr_operations=%" PRId64
1694 "\n",
1695 qdict_get_int(qdict, "rd_bytes"),
1696 qdict_get_int(qdict, "wr_bytes"),
1697 qdict_get_int(qdict, "rd_operations"),
1698 qdict_get_int(qdict, "wr_operations"));
1701 void bdrv_stats_print(Monitor *mon, const QObject *data)
1703 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1706 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1708 QObject *res;
1709 QDict *dict;
1711 res = qobject_from_jsonf("{ 'stats': {"
1712 "'rd_bytes': %" PRId64 ","
1713 "'wr_bytes': %" PRId64 ","
1714 "'rd_operations': %" PRId64 ","
1715 "'wr_operations': %" PRId64 ","
1716 "'wr_highest_offset': %" PRId64
1717 "} }",
1718 bs->rd_bytes, bs->wr_bytes,
1719 bs->rd_ops, bs->wr_ops,
1720 bs->wr_highest_sector *
1721 (uint64_t)BDRV_SECTOR_SIZE);
1722 dict = qobject_to_qdict(res);
1724 if (*bs->device_name) {
1725 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1728 if (bs->file) {
1729 QObject *parent = bdrv_info_stats_bs(bs->file);
1730 qdict_put_obj(dict, "parent", parent);
1733 return res;
1736 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1738 QObject *obj;
1739 QList *devices;
1740 BlockDriverState *bs;
1742 devices = qlist_new();
1744 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1745 obj = bdrv_info_stats_bs(bs);
1746 qlist_append_obj(devices, obj);
1749 *ret_data = QOBJECT(devices);
1752 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1754 if (bs->backing_hd && bs->backing_hd->encrypted)
1755 return bs->backing_file;
1756 else if (bs->encrypted)
1757 return bs->filename;
1758 else
1759 return NULL;
1762 void bdrv_get_backing_filename(BlockDriverState *bs,
1763 char *filename, int filename_size)
1765 if (!bs->backing_file) {
1766 pstrcpy(filename, filename_size, "");
1767 } else {
1768 pstrcpy(filename, filename_size, bs->backing_file);
1772 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1773 const uint8_t *buf, int nb_sectors)
1775 BlockDriver *drv = bs->drv;
1776 if (!drv)
1777 return -ENOMEDIUM;
1778 if (!drv->bdrv_write_compressed)
1779 return -ENOTSUP;
1780 if (bdrv_check_request(bs, sector_num, nb_sectors))
1781 return -EIO;
1783 if (bs->dirty_bitmap) {
1784 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1787 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1790 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1792 BlockDriver *drv = bs->drv;
1793 if (!drv)
1794 return -ENOMEDIUM;
1795 if (!drv->bdrv_get_info)
1796 return -ENOTSUP;
1797 memset(bdi, 0, sizeof(*bdi));
1798 return drv->bdrv_get_info(bs, bdi);
1801 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1802 int64_t pos, int size)
1804 BlockDriver *drv = bs->drv;
1805 if (!drv)
1806 return -ENOMEDIUM;
1807 if (drv->bdrv_save_vmstate)
1808 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1809 if (bs->file)
1810 return bdrv_save_vmstate(bs->file, buf, pos, size);
1811 return -ENOTSUP;
1814 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1815 int64_t pos, int size)
1817 BlockDriver *drv = bs->drv;
1818 if (!drv)
1819 return -ENOMEDIUM;
1820 if (drv->bdrv_load_vmstate)
1821 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1822 if (bs->file)
1823 return bdrv_load_vmstate(bs->file, buf, pos, size);
1824 return -ENOTSUP;
1827 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1829 BlockDriver *drv = bs->drv;
1831 if (!drv || !drv->bdrv_debug_event) {
1832 return;
1835 return drv->bdrv_debug_event(bs, event);
1839 /**************************************************************/
1840 /* handling of snapshots */
1842 int bdrv_can_snapshot(BlockDriverState *bs)
1844 BlockDriver *drv = bs->drv;
1845 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1846 return 0;
1849 if (!drv->bdrv_snapshot_create) {
1850 if (bs->file != NULL) {
1851 return bdrv_can_snapshot(bs->file);
1853 return 0;
1856 return 1;
1859 int bdrv_is_snapshot(BlockDriverState *bs)
1861 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1864 BlockDriverState *bdrv_snapshots(void)
1866 BlockDriverState *bs;
1868 if (bs_snapshots) {
1869 return bs_snapshots;
1872 bs = NULL;
1873 while ((bs = bdrv_next(bs))) {
1874 if (bdrv_can_snapshot(bs)) {
1875 bs_snapshots = bs;
1876 return bs;
1879 return NULL;
1882 int bdrv_snapshot_create(BlockDriverState *bs,
1883 QEMUSnapshotInfo *sn_info)
1885 BlockDriver *drv = bs->drv;
1886 if (!drv)
1887 return -ENOMEDIUM;
1888 if (drv->bdrv_snapshot_create)
1889 return drv->bdrv_snapshot_create(bs, sn_info);
1890 if (bs->file)
1891 return bdrv_snapshot_create(bs->file, sn_info);
1892 return -ENOTSUP;
1895 int bdrv_snapshot_goto(BlockDriverState *bs,
1896 const char *snapshot_id)
1898 BlockDriver *drv = bs->drv;
1899 int ret, open_ret;
1901 if (!drv)
1902 return -ENOMEDIUM;
1903 if (drv->bdrv_snapshot_goto)
1904 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1906 if (bs->file) {
1907 drv->bdrv_close(bs);
1908 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1909 open_ret = drv->bdrv_open(bs, bs->open_flags);
1910 if (open_ret < 0) {
1911 bdrv_delete(bs->file);
1912 bs->drv = NULL;
1913 return open_ret;
1915 return ret;
1918 return -ENOTSUP;
1921 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1923 BlockDriver *drv = bs->drv;
1924 if (!drv)
1925 return -ENOMEDIUM;
1926 if (drv->bdrv_snapshot_delete)
1927 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1928 if (bs->file)
1929 return bdrv_snapshot_delete(bs->file, snapshot_id);
1930 return -ENOTSUP;
1933 int bdrv_snapshot_list(BlockDriverState *bs,
1934 QEMUSnapshotInfo **psn_info)
1936 BlockDriver *drv = bs->drv;
1937 if (!drv)
1938 return -ENOMEDIUM;
1939 if (drv->bdrv_snapshot_list)
1940 return drv->bdrv_snapshot_list(bs, psn_info);
1941 if (bs->file)
1942 return bdrv_snapshot_list(bs->file, psn_info);
1943 return -ENOTSUP;
1946 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
1947 const char *snapshot_name)
1949 BlockDriver *drv = bs->drv;
1950 if (!drv) {
1951 return -ENOMEDIUM;
1953 if (!bs->read_only) {
1954 return -EINVAL;
1956 if (drv->bdrv_snapshot_load_tmp) {
1957 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
1959 return -ENOTSUP;
1962 #define NB_SUFFIXES 4
1964 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1966 static const char suffixes[NB_SUFFIXES] = "KMGT";
1967 int64_t base;
1968 int i;
1970 if (size <= 999) {
1971 snprintf(buf, buf_size, "%" PRId64, size);
1972 } else {
1973 base = 1024;
1974 for(i = 0; i < NB_SUFFIXES; i++) {
1975 if (size < (10 * base)) {
1976 snprintf(buf, buf_size, "%0.1f%c",
1977 (double)size / base,
1978 suffixes[i]);
1979 break;
1980 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1981 snprintf(buf, buf_size, "%" PRId64 "%c",
1982 ((size + (base >> 1)) / base),
1983 suffixes[i]);
1984 break;
1986 base = base * 1024;
1989 return buf;
1992 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1994 char buf1[128], date_buf[128], clock_buf[128];
1995 #ifdef _WIN32
1996 struct tm *ptm;
1997 #else
1998 struct tm tm;
1999 #endif
2000 time_t ti;
2001 int64_t secs;
2003 if (!sn) {
2004 snprintf(buf, buf_size,
2005 "%-10s%-20s%7s%20s%15s",
2006 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2007 } else {
2008 ti = sn->date_sec;
2009 #ifdef _WIN32
2010 ptm = localtime(&ti);
2011 strftime(date_buf, sizeof(date_buf),
2012 "%Y-%m-%d %H:%M:%S", ptm);
2013 #else
2014 localtime_r(&ti, &tm);
2015 strftime(date_buf, sizeof(date_buf),
2016 "%Y-%m-%d %H:%M:%S", &tm);
2017 #endif
2018 secs = sn->vm_clock_nsec / 1000000000;
2019 snprintf(clock_buf, sizeof(clock_buf),
2020 "%02d:%02d:%02d.%03d",
2021 (int)(secs / 3600),
2022 (int)((secs / 60) % 60),
2023 (int)(secs % 60),
2024 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2025 snprintf(buf, buf_size,
2026 "%-10s%-20s%7s%20s%15s",
2027 sn->id_str, sn->name,
2028 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2029 date_buf,
2030 clock_buf);
2032 return buf;
2036 /**************************************************************/
2037 /* async I/Os */
2039 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2040 QEMUIOVector *qiov, int nb_sectors,
2041 BlockDriverCompletionFunc *cb, void *opaque)
2043 BlockDriver *drv = bs->drv;
2044 BlockDriverAIOCB *ret;
2046 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2048 if (!drv)
2049 return NULL;
2050 if (bdrv_check_request(bs, sector_num, nb_sectors))
2051 return NULL;
2053 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2054 cb, opaque);
2056 if (ret) {
2057 /* Update stats even though technically transfer has not happened. */
2058 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2059 bs->rd_ops ++;
2062 return ret;
2065 typedef struct BlockCompleteData {
2066 BlockDriverCompletionFunc *cb;
2067 void *opaque;
2068 BlockDriverState *bs;
2069 int64_t sector_num;
2070 int nb_sectors;
2071 } BlockCompleteData;
2073 static void block_complete_cb(void *opaque, int ret)
2075 BlockCompleteData *b = opaque;
2077 if (b->bs->dirty_bitmap) {
2078 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2080 b->cb(b->opaque, ret);
2081 qemu_free(b);
2084 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2085 int64_t sector_num,
2086 int nb_sectors,
2087 BlockDriverCompletionFunc *cb,
2088 void *opaque)
2090 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2092 blkdata->bs = bs;
2093 blkdata->cb = cb;
2094 blkdata->opaque = opaque;
2095 blkdata->sector_num = sector_num;
2096 blkdata->nb_sectors = nb_sectors;
2098 return blkdata;
2101 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2102 QEMUIOVector *qiov, int nb_sectors,
2103 BlockDriverCompletionFunc *cb, void *opaque)
2105 BlockDriver *drv = bs->drv;
2106 BlockDriverAIOCB *ret;
2107 BlockCompleteData *blk_cb_data;
2109 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2111 if (!drv)
2112 return NULL;
2113 if (bs->read_only)
2114 return NULL;
2115 if (bdrv_check_request(bs, sector_num, nb_sectors))
2116 return NULL;
2118 if (bs->dirty_bitmap) {
2119 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2120 opaque);
2121 cb = &block_complete_cb;
2122 opaque = blk_cb_data;
2125 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2126 cb, opaque);
2128 if (ret) {
2129 /* Update stats even though technically transfer has not happened. */
2130 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2131 bs->wr_ops ++;
2132 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2133 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2137 return ret;
2141 typedef struct MultiwriteCB {
2142 int error;
2143 int num_requests;
2144 int num_callbacks;
2145 struct {
2146 BlockDriverCompletionFunc *cb;
2147 void *opaque;
2148 QEMUIOVector *free_qiov;
2149 void *free_buf;
2150 } callbacks[];
2151 } MultiwriteCB;
2153 static void multiwrite_user_cb(MultiwriteCB *mcb)
2155 int i;
2157 for (i = 0; i < mcb->num_callbacks; i++) {
2158 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2159 if (mcb->callbacks[i].free_qiov) {
2160 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2162 qemu_free(mcb->callbacks[i].free_qiov);
2163 qemu_vfree(mcb->callbacks[i].free_buf);
2167 static void multiwrite_cb(void *opaque, int ret)
2169 MultiwriteCB *mcb = opaque;
2171 trace_multiwrite_cb(mcb, ret);
2173 if (ret < 0 && !mcb->error) {
2174 mcb->error = ret;
2177 mcb->num_requests--;
2178 if (mcb->num_requests == 0) {
2179 multiwrite_user_cb(mcb);
2180 qemu_free(mcb);
2184 static int multiwrite_req_compare(const void *a, const void *b)
2186 const BlockRequest *req1 = a, *req2 = b;
2189 * Note that we can't simply subtract req2->sector from req1->sector
2190 * here as that could overflow the return value.
2192 if (req1->sector > req2->sector) {
2193 return 1;
2194 } else if (req1->sector < req2->sector) {
2195 return -1;
2196 } else {
2197 return 0;
2202 * Takes a bunch of requests and tries to merge them. Returns the number of
2203 * requests that remain after merging.
2205 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2206 int num_reqs, MultiwriteCB *mcb)
2208 int i, outidx;
2210 // Sort requests by start sector
2211 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2213 // Check if adjacent requests touch the same clusters. If so, combine them,
2214 // filling up gaps with zero sectors.
2215 outidx = 0;
2216 for (i = 1; i < num_reqs; i++) {
2217 int merge = 0;
2218 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2220 // This handles the cases that are valid for all block drivers, namely
2221 // exactly sequential writes and overlapping writes.
2222 if (reqs[i].sector <= oldreq_last) {
2223 merge = 1;
2226 // The block driver may decide that it makes sense to combine requests
2227 // even if there is a gap of some sectors between them. In this case,
2228 // the gap is filled with zeros (therefore only applicable for yet
2229 // unused space in format like qcow2).
2230 if (!merge && bs->drv->bdrv_merge_requests) {
2231 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2234 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2235 merge = 0;
2238 if (merge) {
2239 size_t size;
2240 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2241 qemu_iovec_init(qiov,
2242 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2244 // Add the first request to the merged one. If the requests are
2245 // overlapping, drop the last sectors of the first request.
2246 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2247 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2249 // We might need to add some zeros between the two requests
2250 if (reqs[i].sector > oldreq_last) {
2251 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2252 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2253 memset(buf, 0, zero_bytes);
2254 qemu_iovec_add(qiov, buf, zero_bytes);
2255 mcb->callbacks[i].free_buf = buf;
2258 // Add the second request
2259 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2261 reqs[outidx].nb_sectors = qiov->size >> 9;
2262 reqs[outidx].qiov = qiov;
2264 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2265 } else {
2266 outidx++;
2267 reqs[outidx].sector = reqs[i].sector;
2268 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2269 reqs[outidx].qiov = reqs[i].qiov;
2273 return outidx + 1;
2277 * Submit multiple AIO write requests at once.
2279 * On success, the function returns 0 and all requests in the reqs array have
2280 * been submitted. In error case this function returns -1, and any of the
2281 * requests may or may not be submitted yet. In particular, this means that the
2282 * callback will be called for some of the requests, for others it won't. The
2283 * caller must check the error field of the BlockRequest to wait for the right
2284 * callbacks (if error != 0, no callback will be called).
2286 * The implementation may modify the contents of the reqs array, e.g. to merge
2287 * requests. However, the fields opaque and error are left unmodified as they
2288 * are used to signal failure for a single request to the caller.
2290 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2292 BlockDriverAIOCB *acb;
2293 MultiwriteCB *mcb;
2294 int i;
2296 if (num_reqs == 0) {
2297 return 0;
2300 // Create MultiwriteCB structure
2301 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2302 mcb->num_requests = 0;
2303 mcb->num_callbacks = num_reqs;
2305 for (i = 0; i < num_reqs; i++) {
2306 mcb->callbacks[i].cb = reqs[i].cb;
2307 mcb->callbacks[i].opaque = reqs[i].opaque;
2310 // Check for mergable requests
2311 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2313 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2316 * Run the aio requests. As soon as one request can't be submitted
2317 * successfully, fail all requests that are not yet submitted (we must
2318 * return failure for all requests anyway)
2320 * num_requests cannot be set to the right value immediately: If
2321 * bdrv_aio_writev fails for some request, num_requests would be too high
2322 * and therefore multiwrite_cb() would never recognize the multiwrite
2323 * request as completed. We also cannot use the loop variable i to set it
2324 * when the first request fails because the callback may already have been
2325 * called for previously submitted requests. Thus, num_requests must be
2326 * incremented for each request that is submitted.
2328 * The problem that callbacks may be called early also means that we need
2329 * to take care that num_requests doesn't become 0 before all requests are
2330 * submitted - multiwrite_cb() would consider the multiwrite request
2331 * completed. A dummy request that is "completed" by a manual call to
2332 * multiwrite_cb() takes care of this.
2334 mcb->num_requests = 1;
2336 // Run the aio requests
2337 for (i = 0; i < num_reqs; i++) {
2338 mcb->num_requests++;
2339 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2340 reqs[i].nb_sectors, multiwrite_cb, mcb);
2342 if (acb == NULL) {
2343 // We can only fail the whole thing if no request has been
2344 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2345 // complete and report the error in the callback.
2346 if (i == 0) {
2347 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2348 goto fail;
2349 } else {
2350 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2351 multiwrite_cb(mcb, -EIO);
2352 break;
2357 /* Complete the dummy request */
2358 multiwrite_cb(mcb, 0);
2360 return 0;
2362 fail:
2363 for (i = 0; i < mcb->num_callbacks; i++) {
2364 reqs[i].error = -EIO;
2366 qemu_free(mcb);
2367 return -1;
2370 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2371 BlockDriverCompletionFunc *cb, void *opaque)
2373 BlockDriver *drv = bs->drv;
2375 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2376 return bdrv_aio_noop_em(bs, cb, opaque);
2379 if (!drv)
2380 return NULL;
2381 return drv->bdrv_aio_flush(bs, cb, opaque);
2384 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2386 acb->pool->cancel(acb);
2390 /**************************************************************/
2391 /* async block device emulation */
2393 typedef struct BlockDriverAIOCBSync {
2394 BlockDriverAIOCB common;
2395 QEMUBH *bh;
2396 int ret;
2397 /* vector translation state */
2398 QEMUIOVector *qiov;
2399 uint8_t *bounce;
2400 int is_write;
2401 } BlockDriverAIOCBSync;
2403 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2405 BlockDriverAIOCBSync *acb =
2406 container_of(blockacb, BlockDriverAIOCBSync, common);
2407 qemu_bh_delete(acb->bh);
2408 acb->bh = NULL;
2409 qemu_aio_release(acb);
2412 static AIOPool bdrv_em_aio_pool = {
2413 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2414 .cancel = bdrv_aio_cancel_em,
2417 static void bdrv_aio_bh_cb(void *opaque)
2419 BlockDriverAIOCBSync *acb = opaque;
2421 if (!acb->is_write)
2422 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2423 qemu_vfree(acb->bounce);
2424 acb->common.cb(acb->common.opaque, acb->ret);
2425 qemu_bh_delete(acb->bh);
2426 acb->bh = NULL;
2427 qemu_aio_release(acb);
2430 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2431 int64_t sector_num,
2432 QEMUIOVector *qiov,
2433 int nb_sectors,
2434 BlockDriverCompletionFunc *cb,
2435 void *opaque,
2436 int is_write)
2439 BlockDriverAIOCBSync *acb;
2441 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2442 acb->is_write = is_write;
2443 acb->qiov = qiov;
2444 acb->bounce = qemu_blockalign(bs, qiov->size);
2446 if (!acb->bh)
2447 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2449 if (is_write) {
2450 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2451 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2452 } else {
2453 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2456 qemu_bh_schedule(acb->bh);
2458 return &acb->common;
2461 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2462 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2463 BlockDriverCompletionFunc *cb, void *opaque)
2465 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2468 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2469 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2470 BlockDriverCompletionFunc *cb, void *opaque)
2472 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2475 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2476 BlockDriverCompletionFunc *cb, void *opaque)
2478 BlockDriverAIOCBSync *acb;
2480 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2481 acb->is_write = 1; /* don't bounce in the completion hadler */
2482 acb->qiov = NULL;
2483 acb->bounce = NULL;
2484 acb->ret = 0;
2486 if (!acb->bh)
2487 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2489 bdrv_flush(bs);
2490 qemu_bh_schedule(acb->bh);
2491 return &acb->common;
2494 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2495 BlockDriverCompletionFunc *cb, void *opaque)
2497 BlockDriverAIOCBSync *acb;
2499 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2500 acb->is_write = 1; /* don't bounce in the completion handler */
2501 acb->qiov = NULL;
2502 acb->bounce = NULL;
2503 acb->ret = 0;
2505 if (!acb->bh) {
2506 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2509 qemu_bh_schedule(acb->bh);
2510 return &acb->common;
2513 /**************************************************************/
2514 /* sync block device emulation */
2516 static void bdrv_rw_em_cb(void *opaque, int ret)
2518 *(int *)opaque = ret;
2521 #define NOT_DONE 0x7fffffff
2523 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2524 uint8_t *buf, int nb_sectors)
2526 int async_ret;
2527 BlockDriverAIOCB *acb;
2528 struct iovec iov;
2529 QEMUIOVector qiov;
2531 async_context_push();
2533 async_ret = NOT_DONE;
2534 iov.iov_base = (void *)buf;
2535 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2536 qemu_iovec_init_external(&qiov, &iov, 1);
2537 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2538 bdrv_rw_em_cb, &async_ret);
2539 if (acb == NULL) {
2540 async_ret = -1;
2541 goto fail;
2544 while (async_ret == NOT_DONE) {
2545 qemu_aio_wait();
2549 fail:
2550 async_context_pop();
2551 return async_ret;
2554 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2555 const uint8_t *buf, int nb_sectors)
2557 int async_ret;
2558 BlockDriverAIOCB *acb;
2559 struct iovec iov;
2560 QEMUIOVector qiov;
2562 async_context_push();
2564 async_ret = NOT_DONE;
2565 iov.iov_base = (void *)buf;
2566 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2567 qemu_iovec_init_external(&qiov, &iov, 1);
2568 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2569 bdrv_rw_em_cb, &async_ret);
2570 if (acb == NULL) {
2571 async_ret = -1;
2572 goto fail;
2574 while (async_ret == NOT_DONE) {
2575 qemu_aio_wait();
2578 fail:
2579 async_context_pop();
2580 return async_ret;
2583 void bdrv_init(void)
2585 module_call_init(MODULE_INIT_BLOCK);
2588 void bdrv_init_with_whitelist(void)
2590 use_bdrv_whitelist = 1;
2591 bdrv_init();
2594 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2595 BlockDriverCompletionFunc *cb, void *opaque)
2597 BlockDriverAIOCB *acb;
2599 if (pool->free_aiocb) {
2600 acb = pool->free_aiocb;
2601 pool->free_aiocb = acb->next;
2602 } else {
2603 acb = qemu_mallocz(pool->aiocb_size);
2604 acb->pool = pool;
2606 acb->bs = bs;
2607 acb->cb = cb;
2608 acb->opaque = opaque;
2609 return acb;
2612 void qemu_aio_release(void *p)
2614 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2615 AIOPool *pool = acb->pool;
2616 acb->next = pool->free_aiocb;
2617 pool->free_aiocb = acb;
2620 /**************************************************************/
2621 /* removable device support */
2624 * Return TRUE if the media is present
2626 int bdrv_is_inserted(BlockDriverState *bs)
2628 BlockDriver *drv = bs->drv;
2629 int ret;
2630 if (!drv)
2631 return 0;
2632 if (!drv->bdrv_is_inserted)
2633 return !bs->tray_open;
2634 ret = drv->bdrv_is_inserted(bs);
2635 return ret;
2639 * Return TRUE if the media changed since the last call to this
2640 * function. It is currently only used for floppy disks
2642 int bdrv_media_changed(BlockDriverState *bs)
2644 BlockDriver *drv = bs->drv;
2645 int ret;
2647 if (!drv || !drv->bdrv_media_changed)
2648 ret = -ENOTSUP;
2649 else
2650 ret = drv->bdrv_media_changed(bs);
2651 if (ret == -ENOTSUP)
2652 ret = bs->media_changed;
2653 bs->media_changed = 0;
2654 return ret;
2658 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2660 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2662 BlockDriver *drv = bs->drv;
2663 int ret;
2665 if (bs->locked) {
2666 return -EBUSY;
2669 if (!drv || !drv->bdrv_eject) {
2670 ret = -ENOTSUP;
2671 } else {
2672 ret = drv->bdrv_eject(bs, eject_flag);
2674 if (ret == -ENOTSUP) {
2675 ret = 0;
2677 if (ret >= 0) {
2678 bs->tray_open = eject_flag;
2681 return ret;
2684 int bdrv_is_locked(BlockDriverState *bs)
2686 return bs->locked;
2690 * Lock or unlock the media (if it is locked, the user won't be able
2691 * to eject it manually).
2693 void bdrv_set_locked(BlockDriverState *bs, int locked)
2695 BlockDriver *drv = bs->drv;
2697 bs->locked = locked;
2698 if (drv && drv->bdrv_set_locked) {
2699 drv->bdrv_set_locked(bs, locked);
2703 /* needed for generic scsi interface */
2705 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2707 BlockDriver *drv = bs->drv;
2709 if (drv && drv->bdrv_ioctl)
2710 return drv->bdrv_ioctl(bs, req, buf);
2711 return -ENOTSUP;
2714 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2715 unsigned long int req, void *buf,
2716 BlockDriverCompletionFunc *cb, void *opaque)
2718 BlockDriver *drv = bs->drv;
2720 if (drv && drv->bdrv_aio_ioctl)
2721 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2722 return NULL;
2727 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2729 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2732 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2734 int64_t bitmap_size;
2736 bs->dirty_count = 0;
2737 if (enable) {
2738 if (!bs->dirty_bitmap) {
2739 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2740 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2741 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2743 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2745 } else {
2746 if (bs->dirty_bitmap) {
2747 qemu_free(bs->dirty_bitmap);
2748 bs->dirty_bitmap = NULL;
2753 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2755 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2757 if (bs->dirty_bitmap &&
2758 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2759 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2760 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2761 } else {
2762 return 0;
2766 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2767 int nr_sectors)
2769 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2772 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2774 return bs->dirty_count;
2777 int bdrv_img_create(const char *filename, const char *fmt,
2778 const char *base_filename, const char *base_fmt,
2779 char *options, uint64_t img_size, int flags)
2781 QEMUOptionParameter *param = NULL, *create_options = NULL;
2782 QEMUOptionParameter *backing_fmt, *backing_file;
2783 BlockDriverState *bs = NULL;
2784 BlockDriver *drv, *proto_drv;
2785 BlockDriver *backing_drv = NULL;
2786 int ret = 0;
2788 /* Find driver and parse its options */
2789 drv = bdrv_find_format(fmt);
2790 if (!drv) {
2791 error_report("Unknown file format '%s'", fmt);
2792 ret = -EINVAL;
2793 goto out;
2796 proto_drv = bdrv_find_protocol(filename);
2797 if (!proto_drv) {
2798 error_report("Unknown protocol '%s'", filename);
2799 ret = -EINVAL;
2800 goto out;
2803 create_options = append_option_parameters(create_options,
2804 drv->create_options);
2805 create_options = append_option_parameters(create_options,
2806 proto_drv->create_options);
2808 /* Create parameter list with default values */
2809 param = parse_option_parameters("", create_options, param);
2811 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
2813 /* Parse -o options */
2814 if (options) {
2815 param = parse_option_parameters(options, create_options, param);
2816 if (param == NULL) {
2817 error_report("Invalid options for file format '%s'.", fmt);
2818 ret = -EINVAL;
2819 goto out;
2823 if (base_filename) {
2824 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
2825 base_filename)) {
2826 error_report("Backing file not supported for file format '%s'",
2827 fmt);
2828 ret = -EINVAL;
2829 goto out;
2833 if (base_fmt) {
2834 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
2835 error_report("Backing file format not supported for file "
2836 "format '%s'", fmt);
2837 ret = -EINVAL;
2838 goto out;
2842 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
2843 if (backing_file && backing_file->value.s) {
2844 if (!strcmp(filename, backing_file->value.s)) {
2845 error_report("Error: Trying to create an image with the "
2846 "same filename as the backing file");
2847 ret = -EINVAL;
2848 goto out;
2852 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
2853 if (backing_fmt && backing_fmt->value.s) {
2854 backing_drv = bdrv_find_format(backing_fmt->value.s);
2855 if (!backing_drv) {
2856 error_report("Unknown backing file format '%s'",
2857 backing_fmt->value.s);
2858 ret = -EINVAL;
2859 goto out;
2863 // The size for the image must always be specified, with one exception:
2864 // If we are using a backing file, we can obtain the size from there
2865 if (get_option_parameter(param, BLOCK_OPT_SIZE)->value.n == -1) {
2866 if (backing_file && backing_file->value.s) {
2867 uint64_t size;
2868 char buf[32];
2870 bs = bdrv_new("");
2872 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
2873 if (ret < 0) {
2874 error_report("Could not open '%s'", backing_file->value.s);
2875 goto out;
2877 bdrv_get_geometry(bs, &size);
2878 size *= 512;
2880 snprintf(buf, sizeof(buf), "%" PRId64, size);
2881 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
2882 } else {
2883 error_report("Image creation needs a size parameter");
2884 ret = -EINVAL;
2885 goto out;
2889 printf("Formatting '%s', fmt=%s ", filename, fmt);
2890 print_option_parameters(param);
2891 puts("");
2893 ret = bdrv_create(drv, filename, param);
2895 if (ret < 0) {
2896 if (ret == -ENOTSUP) {
2897 error_report("Formatting or formatting option not supported for "
2898 "file format '%s'", fmt);
2899 } else if (ret == -EFBIG) {
2900 error_report("The image size is too large for file format '%s'",
2901 fmt);
2902 } else {
2903 error_report("%s: error while creating %s: %s", filename, fmt,
2904 strerror(-ret));
2908 out:
2909 free_option_parameters(create_options);
2910 free_option_parameters(param);
2912 if (bs) {
2913 bdrv_delete(bs);
2916 return ret;