Fix typos in comments and code (occured -> occurred and related)
[qemu.git] / block.c
blobf40371828ab83e1015e359e5af24a717d5ec1f1c
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 writable.
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 /* make a BlockDriverState anonymous by removing from bdrv_state list.
701 Also, NULL terminate the device_name to prevent double remove */
702 void bdrv_make_anon(BlockDriverState *bs)
704 if (bs->device_name[0] != '\0') {
705 QTAILQ_REMOVE(&bdrv_states, bs, list);
707 bs->device_name[0] = '\0';
710 void bdrv_delete(BlockDriverState *bs)
712 assert(!bs->peer);
714 /* remove from list, if necessary */
715 bdrv_make_anon(bs);
717 bdrv_close(bs);
718 if (bs->file != NULL) {
719 bdrv_delete(bs->file);
722 assert(bs != bs_snapshots);
723 qemu_free(bs);
726 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
728 if (bs->peer) {
729 return -EBUSY;
731 bs->peer = qdev;
732 return 0;
735 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
737 assert(bs->peer == qdev);
738 bs->peer = NULL;
741 DeviceState *bdrv_get_attached(BlockDriverState *bs)
743 return bs->peer;
747 * Run consistency checks on an image
749 * Returns 0 if the check could be completed (it doesn't mean that the image is
750 * free of errors) or -errno when an internal error occurred. The results of the
751 * check are stored in res.
753 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
755 if (bs->drv->bdrv_check == NULL) {
756 return -ENOTSUP;
759 memset(res, 0, sizeof(*res));
760 return bs->drv->bdrv_check(bs, res);
763 #define COMMIT_BUF_SECTORS 2048
765 /* commit COW file into the raw image */
766 int bdrv_commit(BlockDriverState *bs)
768 BlockDriver *drv = bs->drv;
769 BlockDriver *backing_drv;
770 int64_t sector, total_sectors;
771 int n, ro, open_flags;
772 int ret = 0, rw_ret = 0;
773 uint8_t *buf;
774 char filename[1024];
775 BlockDriverState *bs_rw, *bs_ro;
777 if (!drv)
778 return -ENOMEDIUM;
780 if (!bs->backing_hd) {
781 return -ENOTSUP;
784 if (bs->backing_hd->keep_read_only) {
785 return -EACCES;
788 backing_drv = bs->backing_hd->drv;
789 ro = bs->backing_hd->read_only;
790 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
791 open_flags = bs->backing_hd->open_flags;
793 if (ro) {
794 /* re-open as RW */
795 bdrv_delete(bs->backing_hd);
796 bs->backing_hd = NULL;
797 bs_rw = bdrv_new("");
798 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
799 backing_drv);
800 if (rw_ret < 0) {
801 bdrv_delete(bs_rw);
802 /* try to re-open read-only */
803 bs_ro = bdrv_new("");
804 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
805 backing_drv);
806 if (ret < 0) {
807 bdrv_delete(bs_ro);
808 /* drive not functional anymore */
809 bs->drv = NULL;
810 return ret;
812 bs->backing_hd = bs_ro;
813 return rw_ret;
815 bs->backing_hd = bs_rw;
818 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
819 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
821 for (sector = 0; sector < total_sectors; sector += n) {
822 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
824 if (bdrv_read(bs, sector, buf, n) != 0) {
825 ret = -EIO;
826 goto ro_cleanup;
829 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
830 ret = -EIO;
831 goto ro_cleanup;
836 if (drv->bdrv_make_empty) {
837 ret = drv->bdrv_make_empty(bs);
838 bdrv_flush(bs);
842 * Make sure all data we wrote to the backing device is actually
843 * stable on disk.
845 if (bs->backing_hd)
846 bdrv_flush(bs->backing_hd);
848 ro_cleanup:
849 qemu_free(buf);
851 if (ro) {
852 /* re-open as RO */
853 bdrv_delete(bs->backing_hd);
854 bs->backing_hd = NULL;
855 bs_ro = bdrv_new("");
856 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
857 backing_drv);
858 if (ret < 0) {
859 bdrv_delete(bs_ro);
860 /* drive not functional anymore */
861 bs->drv = NULL;
862 return ret;
864 bs->backing_hd = bs_ro;
865 bs->backing_hd->keep_read_only = 0;
868 return ret;
871 void bdrv_commit_all(void)
873 BlockDriverState *bs;
875 QTAILQ_FOREACH(bs, &bdrv_states, list) {
876 bdrv_commit(bs);
881 * Return values:
882 * 0 - success
883 * -EINVAL - backing format specified, but no file
884 * -ENOSPC - can't update the backing file because no space is left in the
885 * image file header
886 * -ENOTSUP - format driver doesn't support changing the backing file
888 int bdrv_change_backing_file(BlockDriverState *bs,
889 const char *backing_file, const char *backing_fmt)
891 BlockDriver *drv = bs->drv;
893 if (drv->bdrv_change_backing_file != NULL) {
894 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
895 } else {
896 return -ENOTSUP;
900 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
901 size_t size)
903 int64_t len;
905 if (!bdrv_is_inserted(bs))
906 return -ENOMEDIUM;
908 if (bs->growable)
909 return 0;
911 len = bdrv_getlength(bs);
913 if (offset < 0)
914 return -EIO;
916 if ((offset > len) || (len - offset < size))
917 return -EIO;
919 return 0;
922 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
923 int nb_sectors)
925 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
926 nb_sectors * BDRV_SECTOR_SIZE);
929 /* return < 0 if error. See bdrv_write() for the return codes */
930 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
931 uint8_t *buf, int nb_sectors)
933 BlockDriver *drv = bs->drv;
935 if (!drv)
936 return -ENOMEDIUM;
937 if (bdrv_check_request(bs, sector_num, nb_sectors))
938 return -EIO;
940 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
943 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
944 int nb_sectors, int dirty)
946 int64_t start, end;
947 unsigned long val, idx, bit;
949 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
950 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
952 for (; start <= end; start++) {
953 idx = start / (sizeof(unsigned long) * 8);
954 bit = start % (sizeof(unsigned long) * 8);
955 val = bs->dirty_bitmap[idx];
956 if (dirty) {
957 if (!(val & (1UL << bit))) {
958 bs->dirty_count++;
959 val |= 1UL << bit;
961 } else {
962 if (val & (1UL << bit)) {
963 bs->dirty_count--;
964 val &= ~(1UL << bit);
967 bs->dirty_bitmap[idx] = val;
971 /* Return < 0 if error. Important errors are:
972 -EIO generic I/O error (may happen for all errors)
973 -ENOMEDIUM No media inserted.
974 -EINVAL Invalid sector number or nb_sectors
975 -EACCES Trying to write a read-only device
977 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
978 const uint8_t *buf, int nb_sectors)
980 BlockDriver *drv = bs->drv;
981 if (!bs->drv)
982 return -ENOMEDIUM;
983 if (bs->read_only)
984 return -EACCES;
985 if (bdrv_check_request(bs, sector_num, nb_sectors))
986 return -EIO;
988 if (bs->dirty_bitmap) {
989 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
992 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
993 bs->wr_highest_sector = sector_num + nb_sectors - 1;
996 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
999 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1000 void *buf, int count1)
1002 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1003 int len, nb_sectors, count;
1004 int64_t sector_num;
1005 int ret;
1007 count = count1;
1008 /* first read to align to sector start */
1009 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1010 if (len > count)
1011 len = count;
1012 sector_num = offset >> BDRV_SECTOR_BITS;
1013 if (len > 0) {
1014 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1015 return ret;
1016 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1017 count -= len;
1018 if (count == 0)
1019 return count1;
1020 sector_num++;
1021 buf += len;
1024 /* read the sectors "in place" */
1025 nb_sectors = count >> BDRV_SECTOR_BITS;
1026 if (nb_sectors > 0) {
1027 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1028 return ret;
1029 sector_num += nb_sectors;
1030 len = nb_sectors << BDRV_SECTOR_BITS;
1031 buf += len;
1032 count -= len;
1035 /* add data from the last sector */
1036 if (count > 0) {
1037 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1038 return ret;
1039 memcpy(buf, tmp_buf, count);
1041 return count1;
1044 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1045 const void *buf, int count1)
1047 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1048 int len, nb_sectors, count;
1049 int64_t sector_num;
1050 int ret;
1052 count = count1;
1053 /* first write to align to sector start */
1054 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1055 if (len > count)
1056 len = count;
1057 sector_num = offset >> BDRV_SECTOR_BITS;
1058 if (len > 0) {
1059 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1060 return ret;
1061 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1062 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1063 return ret;
1064 count -= len;
1065 if (count == 0)
1066 return count1;
1067 sector_num++;
1068 buf += len;
1071 /* write the sectors "in place" */
1072 nb_sectors = count >> BDRV_SECTOR_BITS;
1073 if (nb_sectors > 0) {
1074 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1075 return ret;
1076 sector_num += nb_sectors;
1077 len = nb_sectors << BDRV_SECTOR_BITS;
1078 buf += len;
1079 count -= len;
1082 /* add data from the last sector */
1083 if (count > 0) {
1084 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1085 return ret;
1086 memcpy(tmp_buf, buf, count);
1087 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1088 return ret;
1090 return count1;
1094 * Writes to the file and ensures that no writes are reordered across this
1095 * request (acts as a barrier)
1097 * Returns 0 on success, -errno in error cases.
1099 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1100 const void *buf, int count)
1102 int ret;
1104 ret = bdrv_pwrite(bs, offset, buf, count);
1105 if (ret < 0) {
1106 return ret;
1109 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1110 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1111 bdrv_flush(bs);
1114 return 0;
1118 * Writes to the file and ensures that no writes are reordered across this
1119 * request (acts as a barrier)
1121 * Returns 0 on success, -errno in error cases.
1123 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1124 const uint8_t *buf, int nb_sectors)
1126 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1127 buf, BDRV_SECTOR_SIZE * nb_sectors);
1131 * Truncate file to 'offset' bytes (needed only for file protocols)
1133 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1135 BlockDriver *drv = bs->drv;
1136 int ret;
1137 if (!drv)
1138 return -ENOMEDIUM;
1139 if (!drv->bdrv_truncate)
1140 return -ENOTSUP;
1141 if (bs->read_only)
1142 return -EACCES;
1143 if (bdrv_in_use(bs))
1144 return -EBUSY;
1145 ret = drv->bdrv_truncate(bs, offset);
1146 if (ret == 0) {
1147 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1148 if (bs->change_cb) {
1149 bs->change_cb(bs->change_opaque, CHANGE_SIZE);
1152 return ret;
1156 * Length of a file in bytes. Return < 0 if error or unknown.
1158 int64_t bdrv_getlength(BlockDriverState *bs)
1160 BlockDriver *drv = bs->drv;
1161 if (!drv)
1162 return -ENOMEDIUM;
1164 if (bs->growable || bs->removable) {
1165 if (drv->bdrv_getlength) {
1166 return drv->bdrv_getlength(bs);
1169 return bs->total_sectors * BDRV_SECTOR_SIZE;
1172 /* return 0 as number of sectors if no device present or error */
1173 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1175 int64_t length;
1176 length = bdrv_getlength(bs);
1177 if (length < 0)
1178 length = 0;
1179 else
1180 length = length >> BDRV_SECTOR_BITS;
1181 *nb_sectors_ptr = length;
1184 struct partition {
1185 uint8_t boot_ind; /* 0x80 - active */
1186 uint8_t head; /* starting head */
1187 uint8_t sector; /* starting sector */
1188 uint8_t cyl; /* starting cylinder */
1189 uint8_t sys_ind; /* What partition type */
1190 uint8_t end_head; /* end head */
1191 uint8_t end_sector; /* end sector */
1192 uint8_t end_cyl; /* end cylinder */
1193 uint32_t start_sect; /* starting sector counting from 0 */
1194 uint32_t nr_sects; /* nr of sectors in partition */
1195 } __attribute__((packed));
1197 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1198 static int guess_disk_lchs(BlockDriverState *bs,
1199 int *pcylinders, int *pheads, int *psectors)
1201 uint8_t buf[BDRV_SECTOR_SIZE];
1202 int ret, i, heads, sectors, cylinders;
1203 struct partition *p;
1204 uint32_t nr_sects;
1205 uint64_t nb_sectors;
1207 bdrv_get_geometry(bs, &nb_sectors);
1209 ret = bdrv_read(bs, 0, buf, 1);
1210 if (ret < 0)
1211 return -1;
1212 /* test msdos magic */
1213 if (buf[510] != 0x55 || buf[511] != 0xaa)
1214 return -1;
1215 for(i = 0; i < 4; i++) {
1216 p = ((struct partition *)(buf + 0x1be)) + i;
1217 nr_sects = le32_to_cpu(p->nr_sects);
1218 if (nr_sects && p->end_head) {
1219 /* We make the assumption that the partition terminates on
1220 a cylinder boundary */
1221 heads = p->end_head + 1;
1222 sectors = p->end_sector & 63;
1223 if (sectors == 0)
1224 continue;
1225 cylinders = nb_sectors / (heads * sectors);
1226 if (cylinders < 1 || cylinders > 16383)
1227 continue;
1228 *pheads = heads;
1229 *psectors = sectors;
1230 *pcylinders = cylinders;
1231 #if 0
1232 printf("guessed geometry: LCHS=%d %d %d\n",
1233 cylinders, heads, sectors);
1234 #endif
1235 return 0;
1238 return -1;
1241 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1243 int translation, lba_detected = 0;
1244 int cylinders, heads, secs;
1245 uint64_t nb_sectors;
1247 /* if a geometry hint is available, use it */
1248 bdrv_get_geometry(bs, &nb_sectors);
1249 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1250 translation = bdrv_get_translation_hint(bs);
1251 if (cylinders != 0) {
1252 *pcyls = cylinders;
1253 *pheads = heads;
1254 *psecs = secs;
1255 } else {
1256 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1257 if (heads > 16) {
1258 /* if heads > 16, it means that a BIOS LBA
1259 translation was active, so the default
1260 hardware geometry is OK */
1261 lba_detected = 1;
1262 goto default_geometry;
1263 } else {
1264 *pcyls = cylinders;
1265 *pheads = heads;
1266 *psecs = secs;
1267 /* disable any translation to be in sync with
1268 the logical geometry */
1269 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1270 bdrv_set_translation_hint(bs,
1271 BIOS_ATA_TRANSLATION_NONE);
1274 } else {
1275 default_geometry:
1276 /* if no geometry, use a standard physical disk geometry */
1277 cylinders = nb_sectors / (16 * 63);
1279 if (cylinders > 16383)
1280 cylinders = 16383;
1281 else if (cylinders < 2)
1282 cylinders = 2;
1283 *pcyls = cylinders;
1284 *pheads = 16;
1285 *psecs = 63;
1286 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1287 if ((*pcyls * *pheads) <= 131072) {
1288 bdrv_set_translation_hint(bs,
1289 BIOS_ATA_TRANSLATION_LARGE);
1290 } else {
1291 bdrv_set_translation_hint(bs,
1292 BIOS_ATA_TRANSLATION_LBA);
1296 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1300 void bdrv_set_geometry_hint(BlockDriverState *bs,
1301 int cyls, int heads, int secs)
1303 bs->cyls = cyls;
1304 bs->heads = heads;
1305 bs->secs = secs;
1308 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1310 bs->type = type;
1311 bs->removable = ((type == BDRV_TYPE_CDROM ||
1312 type == BDRV_TYPE_FLOPPY));
1315 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1317 bs->translation = translation;
1320 void bdrv_get_geometry_hint(BlockDriverState *bs,
1321 int *pcyls, int *pheads, int *psecs)
1323 *pcyls = bs->cyls;
1324 *pheads = bs->heads;
1325 *psecs = bs->secs;
1328 /* Recognize floppy formats */
1329 typedef struct FDFormat {
1330 FDriveType drive;
1331 uint8_t last_sect;
1332 uint8_t max_track;
1333 uint8_t max_head;
1334 } FDFormat;
1336 static const FDFormat fd_formats[] = {
1337 /* First entry is default format */
1338 /* 1.44 MB 3"1/2 floppy disks */
1339 { FDRIVE_DRV_144, 18, 80, 1, },
1340 { FDRIVE_DRV_144, 20, 80, 1, },
1341 { FDRIVE_DRV_144, 21, 80, 1, },
1342 { FDRIVE_DRV_144, 21, 82, 1, },
1343 { FDRIVE_DRV_144, 21, 83, 1, },
1344 { FDRIVE_DRV_144, 22, 80, 1, },
1345 { FDRIVE_DRV_144, 23, 80, 1, },
1346 { FDRIVE_DRV_144, 24, 80, 1, },
1347 /* 2.88 MB 3"1/2 floppy disks */
1348 { FDRIVE_DRV_288, 36, 80, 1, },
1349 { FDRIVE_DRV_288, 39, 80, 1, },
1350 { FDRIVE_DRV_288, 40, 80, 1, },
1351 { FDRIVE_DRV_288, 44, 80, 1, },
1352 { FDRIVE_DRV_288, 48, 80, 1, },
1353 /* 720 kB 3"1/2 floppy disks */
1354 { FDRIVE_DRV_144, 9, 80, 1, },
1355 { FDRIVE_DRV_144, 10, 80, 1, },
1356 { FDRIVE_DRV_144, 10, 82, 1, },
1357 { FDRIVE_DRV_144, 10, 83, 1, },
1358 { FDRIVE_DRV_144, 13, 80, 1, },
1359 { FDRIVE_DRV_144, 14, 80, 1, },
1360 /* 1.2 MB 5"1/4 floppy disks */
1361 { FDRIVE_DRV_120, 15, 80, 1, },
1362 { FDRIVE_DRV_120, 18, 80, 1, },
1363 { FDRIVE_DRV_120, 18, 82, 1, },
1364 { FDRIVE_DRV_120, 18, 83, 1, },
1365 { FDRIVE_DRV_120, 20, 80, 1, },
1366 /* 720 kB 5"1/4 floppy disks */
1367 { FDRIVE_DRV_120, 9, 80, 1, },
1368 { FDRIVE_DRV_120, 11, 80, 1, },
1369 /* 360 kB 5"1/4 floppy disks */
1370 { FDRIVE_DRV_120, 9, 40, 1, },
1371 { FDRIVE_DRV_120, 9, 40, 0, },
1372 { FDRIVE_DRV_120, 10, 41, 1, },
1373 { FDRIVE_DRV_120, 10, 42, 1, },
1374 /* 320 kB 5"1/4 floppy disks */
1375 { FDRIVE_DRV_120, 8, 40, 1, },
1376 { FDRIVE_DRV_120, 8, 40, 0, },
1377 /* 360 kB must match 5"1/4 better than 3"1/2... */
1378 { FDRIVE_DRV_144, 9, 80, 0, },
1379 /* end */
1380 { FDRIVE_DRV_NONE, -1, -1, 0, },
1383 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
1384 int *max_track, int *last_sect,
1385 FDriveType drive_in, FDriveType *drive)
1387 const FDFormat *parse;
1388 uint64_t nb_sectors, size;
1389 int i, first_match, match;
1391 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
1392 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
1393 /* User defined disk */
1394 } else {
1395 bdrv_get_geometry(bs, &nb_sectors);
1396 match = -1;
1397 first_match = -1;
1398 for (i = 0; ; i++) {
1399 parse = &fd_formats[i];
1400 if (parse->drive == FDRIVE_DRV_NONE) {
1401 break;
1403 if (drive_in == parse->drive ||
1404 drive_in == FDRIVE_DRV_NONE) {
1405 size = (parse->max_head + 1) * parse->max_track *
1406 parse->last_sect;
1407 if (nb_sectors == size) {
1408 match = i;
1409 break;
1411 if (first_match == -1) {
1412 first_match = i;
1416 if (match == -1) {
1417 if (first_match == -1) {
1418 match = 1;
1419 } else {
1420 match = first_match;
1422 parse = &fd_formats[match];
1424 *nb_heads = parse->max_head + 1;
1425 *max_track = parse->max_track;
1426 *last_sect = parse->last_sect;
1427 *drive = parse->drive;
1431 int bdrv_get_type_hint(BlockDriverState *bs)
1433 return bs->type;
1436 int bdrv_get_translation_hint(BlockDriverState *bs)
1438 return bs->translation;
1441 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1442 BlockErrorAction on_write_error)
1444 bs->on_read_error = on_read_error;
1445 bs->on_write_error = on_write_error;
1448 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1450 return is_read ? bs->on_read_error : bs->on_write_error;
1453 void bdrv_set_removable(BlockDriverState *bs, int removable)
1455 bs->removable = removable;
1456 if (removable && bs == bs_snapshots) {
1457 bs_snapshots = NULL;
1461 int bdrv_is_removable(BlockDriverState *bs)
1463 return bs->removable;
1466 int bdrv_is_read_only(BlockDriverState *bs)
1468 return bs->read_only;
1471 int bdrv_is_sg(BlockDriverState *bs)
1473 return bs->sg;
1476 int bdrv_enable_write_cache(BlockDriverState *bs)
1478 return bs->enable_write_cache;
1481 /* XXX: no longer used */
1482 void bdrv_set_change_cb(BlockDriverState *bs,
1483 void (*change_cb)(void *opaque, int reason),
1484 void *opaque)
1486 bs->change_cb = change_cb;
1487 bs->change_opaque = opaque;
1490 int bdrv_is_encrypted(BlockDriverState *bs)
1492 if (bs->backing_hd && bs->backing_hd->encrypted)
1493 return 1;
1494 return bs->encrypted;
1497 int bdrv_key_required(BlockDriverState *bs)
1499 BlockDriverState *backing_hd = bs->backing_hd;
1501 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1502 return 1;
1503 return (bs->encrypted && !bs->valid_key);
1506 int bdrv_set_key(BlockDriverState *bs, const char *key)
1508 int ret;
1509 if (bs->backing_hd && bs->backing_hd->encrypted) {
1510 ret = bdrv_set_key(bs->backing_hd, key);
1511 if (ret < 0)
1512 return ret;
1513 if (!bs->encrypted)
1514 return 0;
1516 if (!bs->encrypted) {
1517 return -EINVAL;
1518 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1519 return -ENOMEDIUM;
1521 ret = bs->drv->bdrv_set_key(bs, key);
1522 if (ret < 0) {
1523 bs->valid_key = 0;
1524 } else if (!bs->valid_key) {
1525 bs->valid_key = 1;
1526 /* call the change callback now, we skipped it on open */
1527 bs->media_changed = 1;
1528 if (bs->change_cb)
1529 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
1531 return ret;
1534 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1536 if (!bs->drv) {
1537 buf[0] = '\0';
1538 } else {
1539 pstrcpy(buf, buf_size, bs->drv->format_name);
1543 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1544 void *opaque)
1546 BlockDriver *drv;
1548 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1549 it(opaque, drv->format_name);
1553 BlockDriverState *bdrv_find(const char *name)
1555 BlockDriverState *bs;
1557 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1558 if (!strcmp(name, bs->device_name)) {
1559 return bs;
1562 return NULL;
1565 BlockDriverState *bdrv_next(BlockDriverState *bs)
1567 if (!bs) {
1568 return QTAILQ_FIRST(&bdrv_states);
1570 return QTAILQ_NEXT(bs, list);
1573 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1575 BlockDriverState *bs;
1577 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1578 it(opaque, bs);
1582 const char *bdrv_get_device_name(BlockDriverState *bs)
1584 return bs->device_name;
1587 int bdrv_flush(BlockDriverState *bs)
1589 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1590 return 0;
1593 if (bs->drv && bs->drv->bdrv_flush) {
1594 return bs->drv->bdrv_flush(bs);
1598 * Some block drivers always operate in either writethrough or unsafe mode
1599 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1600 * the server works (because the behaviour is hardcoded or depends on
1601 * server-side configuration), so we can't ensure that everything is safe
1602 * on disk. Returning an error doesn't work because that would break guests
1603 * even if the server operates in writethrough mode.
1605 * Let's hope the user knows what he's doing.
1607 return 0;
1610 void bdrv_flush_all(void)
1612 BlockDriverState *bs;
1614 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1615 if (bs->drv && !bdrv_is_read_only(bs) &&
1616 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1617 bdrv_flush(bs);
1622 int bdrv_has_zero_init(BlockDriverState *bs)
1624 assert(bs->drv);
1626 if (bs->drv->bdrv_has_zero_init) {
1627 return bs->drv->bdrv_has_zero_init(bs);
1630 return 1;
1633 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1635 if (!bs->drv) {
1636 return -ENOMEDIUM;
1638 if (!bs->drv->bdrv_discard) {
1639 return 0;
1641 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1645 * Returns true iff the specified sector is present in the disk image. Drivers
1646 * not implementing the functionality are assumed to not support backing files,
1647 * hence all their sectors are reported as allocated.
1649 * 'pnum' is set to the number of sectors (including and immediately following
1650 * the specified sector) that are known to be in the same
1651 * allocated/unallocated state.
1653 * 'nb_sectors' is the max value 'pnum' should be set to.
1655 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1656 int *pnum)
1658 int64_t n;
1659 if (!bs->drv->bdrv_is_allocated) {
1660 if (sector_num >= bs->total_sectors) {
1661 *pnum = 0;
1662 return 0;
1664 n = bs->total_sectors - sector_num;
1665 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1666 return 1;
1668 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1671 void bdrv_mon_event(const BlockDriverState *bdrv,
1672 BlockMonEventAction action, int is_read)
1674 QObject *data;
1675 const char *action_str;
1677 switch (action) {
1678 case BDRV_ACTION_REPORT:
1679 action_str = "report";
1680 break;
1681 case BDRV_ACTION_IGNORE:
1682 action_str = "ignore";
1683 break;
1684 case BDRV_ACTION_STOP:
1685 action_str = "stop";
1686 break;
1687 default:
1688 abort();
1691 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1692 bdrv->device_name,
1693 action_str,
1694 is_read ? "read" : "write");
1695 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1697 qobject_decref(data);
1700 static void bdrv_print_dict(QObject *obj, void *opaque)
1702 QDict *bs_dict;
1703 Monitor *mon = opaque;
1705 bs_dict = qobject_to_qdict(obj);
1707 monitor_printf(mon, "%s: type=%s removable=%d",
1708 qdict_get_str(bs_dict, "device"),
1709 qdict_get_str(bs_dict, "type"),
1710 qdict_get_bool(bs_dict, "removable"));
1712 if (qdict_get_bool(bs_dict, "removable")) {
1713 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1716 if (qdict_haskey(bs_dict, "inserted")) {
1717 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1719 monitor_printf(mon, " file=");
1720 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1721 if (qdict_haskey(qdict, "backing_file")) {
1722 monitor_printf(mon, " backing_file=");
1723 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1725 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1726 qdict_get_bool(qdict, "ro"),
1727 qdict_get_str(qdict, "drv"),
1728 qdict_get_bool(qdict, "encrypted"));
1729 } else {
1730 monitor_printf(mon, " [not inserted]");
1733 monitor_printf(mon, "\n");
1736 void bdrv_info_print(Monitor *mon, const QObject *data)
1738 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1741 void bdrv_info(Monitor *mon, QObject **ret_data)
1743 QList *bs_list;
1744 BlockDriverState *bs;
1746 bs_list = qlist_new();
1748 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1749 QObject *bs_obj;
1750 const char *type = "unknown";
1752 switch(bs->type) {
1753 case BDRV_TYPE_HD:
1754 type = "hd";
1755 break;
1756 case BDRV_TYPE_CDROM:
1757 type = "cdrom";
1758 break;
1759 case BDRV_TYPE_FLOPPY:
1760 type = "floppy";
1761 break;
1764 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1765 "'removable': %i, 'locked': %i }",
1766 bs->device_name, type, bs->removable,
1767 bs->locked);
1769 if (bs->drv) {
1770 QObject *obj;
1771 QDict *bs_dict = qobject_to_qdict(bs_obj);
1773 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1774 "'encrypted': %i }",
1775 bs->filename, bs->read_only,
1776 bs->drv->format_name,
1777 bdrv_is_encrypted(bs));
1778 if (bs->backing_file[0] != '\0') {
1779 QDict *qdict = qobject_to_qdict(obj);
1780 qdict_put(qdict, "backing_file",
1781 qstring_from_str(bs->backing_file));
1784 qdict_put_obj(bs_dict, "inserted", obj);
1786 qlist_append_obj(bs_list, bs_obj);
1789 *ret_data = QOBJECT(bs_list);
1792 static void bdrv_stats_iter(QObject *data, void *opaque)
1794 QDict *qdict;
1795 Monitor *mon = opaque;
1797 qdict = qobject_to_qdict(data);
1798 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1800 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1801 monitor_printf(mon, " rd_bytes=%" PRId64
1802 " wr_bytes=%" PRId64
1803 " rd_operations=%" PRId64
1804 " wr_operations=%" PRId64
1805 "\n",
1806 qdict_get_int(qdict, "rd_bytes"),
1807 qdict_get_int(qdict, "wr_bytes"),
1808 qdict_get_int(qdict, "rd_operations"),
1809 qdict_get_int(qdict, "wr_operations"));
1812 void bdrv_stats_print(Monitor *mon, const QObject *data)
1814 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1817 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1819 QObject *res;
1820 QDict *dict;
1822 res = qobject_from_jsonf("{ 'stats': {"
1823 "'rd_bytes': %" PRId64 ","
1824 "'wr_bytes': %" PRId64 ","
1825 "'rd_operations': %" PRId64 ","
1826 "'wr_operations': %" PRId64 ","
1827 "'wr_highest_offset': %" PRId64
1828 "} }",
1829 bs->rd_bytes, bs->wr_bytes,
1830 bs->rd_ops, bs->wr_ops,
1831 bs->wr_highest_sector *
1832 (uint64_t)BDRV_SECTOR_SIZE);
1833 dict = qobject_to_qdict(res);
1835 if (*bs->device_name) {
1836 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1839 if (bs->file) {
1840 QObject *parent = bdrv_info_stats_bs(bs->file);
1841 qdict_put_obj(dict, "parent", parent);
1844 return res;
1847 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1849 QObject *obj;
1850 QList *devices;
1851 BlockDriverState *bs;
1853 devices = qlist_new();
1855 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1856 obj = bdrv_info_stats_bs(bs);
1857 qlist_append_obj(devices, obj);
1860 *ret_data = QOBJECT(devices);
1863 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1865 if (bs->backing_hd && bs->backing_hd->encrypted)
1866 return bs->backing_file;
1867 else if (bs->encrypted)
1868 return bs->filename;
1869 else
1870 return NULL;
1873 void bdrv_get_backing_filename(BlockDriverState *bs,
1874 char *filename, int filename_size)
1876 if (!bs->backing_file) {
1877 pstrcpy(filename, filename_size, "");
1878 } else {
1879 pstrcpy(filename, filename_size, bs->backing_file);
1883 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1884 const uint8_t *buf, int nb_sectors)
1886 BlockDriver *drv = bs->drv;
1887 if (!drv)
1888 return -ENOMEDIUM;
1889 if (!drv->bdrv_write_compressed)
1890 return -ENOTSUP;
1891 if (bdrv_check_request(bs, sector_num, nb_sectors))
1892 return -EIO;
1894 if (bs->dirty_bitmap) {
1895 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1898 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1901 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1903 BlockDriver *drv = bs->drv;
1904 if (!drv)
1905 return -ENOMEDIUM;
1906 if (!drv->bdrv_get_info)
1907 return -ENOTSUP;
1908 memset(bdi, 0, sizeof(*bdi));
1909 return drv->bdrv_get_info(bs, bdi);
1912 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1913 int64_t pos, int size)
1915 BlockDriver *drv = bs->drv;
1916 if (!drv)
1917 return -ENOMEDIUM;
1918 if (drv->bdrv_save_vmstate)
1919 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1920 if (bs->file)
1921 return bdrv_save_vmstate(bs->file, buf, pos, size);
1922 return -ENOTSUP;
1925 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1926 int64_t pos, int size)
1928 BlockDriver *drv = bs->drv;
1929 if (!drv)
1930 return -ENOMEDIUM;
1931 if (drv->bdrv_load_vmstate)
1932 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1933 if (bs->file)
1934 return bdrv_load_vmstate(bs->file, buf, pos, size);
1935 return -ENOTSUP;
1938 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1940 BlockDriver *drv = bs->drv;
1942 if (!drv || !drv->bdrv_debug_event) {
1943 return;
1946 return drv->bdrv_debug_event(bs, event);
1950 /**************************************************************/
1951 /* handling of snapshots */
1953 int bdrv_can_snapshot(BlockDriverState *bs)
1955 BlockDriver *drv = bs->drv;
1956 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1957 return 0;
1960 if (!drv->bdrv_snapshot_create) {
1961 if (bs->file != NULL) {
1962 return bdrv_can_snapshot(bs->file);
1964 return 0;
1967 return 1;
1970 int bdrv_is_snapshot(BlockDriverState *bs)
1972 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1975 BlockDriverState *bdrv_snapshots(void)
1977 BlockDriverState *bs;
1979 if (bs_snapshots) {
1980 return bs_snapshots;
1983 bs = NULL;
1984 while ((bs = bdrv_next(bs))) {
1985 if (bdrv_can_snapshot(bs)) {
1986 bs_snapshots = bs;
1987 return bs;
1990 return NULL;
1993 int bdrv_snapshot_create(BlockDriverState *bs,
1994 QEMUSnapshotInfo *sn_info)
1996 BlockDriver *drv = bs->drv;
1997 if (!drv)
1998 return -ENOMEDIUM;
1999 if (drv->bdrv_snapshot_create)
2000 return drv->bdrv_snapshot_create(bs, sn_info);
2001 if (bs->file)
2002 return bdrv_snapshot_create(bs->file, sn_info);
2003 return -ENOTSUP;
2006 int bdrv_snapshot_goto(BlockDriverState *bs,
2007 const char *snapshot_id)
2009 BlockDriver *drv = bs->drv;
2010 int ret, open_ret;
2012 if (!drv)
2013 return -ENOMEDIUM;
2014 if (drv->bdrv_snapshot_goto)
2015 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2017 if (bs->file) {
2018 drv->bdrv_close(bs);
2019 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2020 open_ret = drv->bdrv_open(bs, bs->open_flags);
2021 if (open_ret < 0) {
2022 bdrv_delete(bs->file);
2023 bs->drv = NULL;
2024 return open_ret;
2026 return ret;
2029 return -ENOTSUP;
2032 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2034 BlockDriver *drv = bs->drv;
2035 if (!drv)
2036 return -ENOMEDIUM;
2037 if (drv->bdrv_snapshot_delete)
2038 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2039 if (bs->file)
2040 return bdrv_snapshot_delete(bs->file, snapshot_id);
2041 return -ENOTSUP;
2044 int bdrv_snapshot_list(BlockDriverState *bs,
2045 QEMUSnapshotInfo **psn_info)
2047 BlockDriver *drv = bs->drv;
2048 if (!drv)
2049 return -ENOMEDIUM;
2050 if (drv->bdrv_snapshot_list)
2051 return drv->bdrv_snapshot_list(bs, psn_info);
2052 if (bs->file)
2053 return bdrv_snapshot_list(bs->file, psn_info);
2054 return -ENOTSUP;
2057 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2058 const char *snapshot_name)
2060 BlockDriver *drv = bs->drv;
2061 if (!drv) {
2062 return -ENOMEDIUM;
2064 if (!bs->read_only) {
2065 return -EINVAL;
2067 if (drv->bdrv_snapshot_load_tmp) {
2068 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2070 return -ENOTSUP;
2073 #define NB_SUFFIXES 4
2075 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2077 static const char suffixes[NB_SUFFIXES] = "KMGT";
2078 int64_t base;
2079 int i;
2081 if (size <= 999) {
2082 snprintf(buf, buf_size, "%" PRId64, size);
2083 } else {
2084 base = 1024;
2085 for(i = 0; i < NB_SUFFIXES; i++) {
2086 if (size < (10 * base)) {
2087 snprintf(buf, buf_size, "%0.1f%c",
2088 (double)size / base,
2089 suffixes[i]);
2090 break;
2091 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2092 snprintf(buf, buf_size, "%" PRId64 "%c",
2093 ((size + (base >> 1)) / base),
2094 suffixes[i]);
2095 break;
2097 base = base * 1024;
2100 return buf;
2103 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2105 char buf1[128], date_buf[128], clock_buf[128];
2106 #ifdef _WIN32
2107 struct tm *ptm;
2108 #else
2109 struct tm tm;
2110 #endif
2111 time_t ti;
2112 int64_t secs;
2114 if (!sn) {
2115 snprintf(buf, buf_size,
2116 "%-10s%-20s%7s%20s%15s",
2117 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2118 } else {
2119 ti = sn->date_sec;
2120 #ifdef _WIN32
2121 ptm = localtime(&ti);
2122 strftime(date_buf, sizeof(date_buf),
2123 "%Y-%m-%d %H:%M:%S", ptm);
2124 #else
2125 localtime_r(&ti, &tm);
2126 strftime(date_buf, sizeof(date_buf),
2127 "%Y-%m-%d %H:%M:%S", &tm);
2128 #endif
2129 secs = sn->vm_clock_nsec / 1000000000;
2130 snprintf(clock_buf, sizeof(clock_buf),
2131 "%02d:%02d:%02d.%03d",
2132 (int)(secs / 3600),
2133 (int)((secs / 60) % 60),
2134 (int)(secs % 60),
2135 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2136 snprintf(buf, buf_size,
2137 "%-10s%-20s%7s%20s%15s",
2138 sn->id_str, sn->name,
2139 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2140 date_buf,
2141 clock_buf);
2143 return buf;
2147 /**************************************************************/
2148 /* async I/Os */
2150 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2151 QEMUIOVector *qiov, int nb_sectors,
2152 BlockDriverCompletionFunc *cb, void *opaque)
2154 BlockDriver *drv = bs->drv;
2155 BlockDriverAIOCB *ret;
2157 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2159 if (!drv)
2160 return NULL;
2161 if (bdrv_check_request(bs, sector_num, nb_sectors))
2162 return NULL;
2164 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2165 cb, opaque);
2167 if (ret) {
2168 /* Update stats even though technically transfer has not happened. */
2169 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2170 bs->rd_ops ++;
2173 return ret;
2176 typedef struct BlockCompleteData {
2177 BlockDriverCompletionFunc *cb;
2178 void *opaque;
2179 BlockDriverState *bs;
2180 int64_t sector_num;
2181 int nb_sectors;
2182 } BlockCompleteData;
2184 static void block_complete_cb(void *opaque, int ret)
2186 BlockCompleteData *b = opaque;
2188 if (b->bs->dirty_bitmap) {
2189 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2191 b->cb(b->opaque, ret);
2192 qemu_free(b);
2195 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2196 int64_t sector_num,
2197 int nb_sectors,
2198 BlockDriverCompletionFunc *cb,
2199 void *opaque)
2201 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2203 blkdata->bs = bs;
2204 blkdata->cb = cb;
2205 blkdata->opaque = opaque;
2206 blkdata->sector_num = sector_num;
2207 blkdata->nb_sectors = nb_sectors;
2209 return blkdata;
2212 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2213 QEMUIOVector *qiov, int nb_sectors,
2214 BlockDriverCompletionFunc *cb, void *opaque)
2216 BlockDriver *drv = bs->drv;
2217 BlockDriverAIOCB *ret;
2218 BlockCompleteData *blk_cb_data;
2220 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2222 if (!drv)
2223 return NULL;
2224 if (bs->read_only)
2225 return NULL;
2226 if (bdrv_check_request(bs, sector_num, nb_sectors))
2227 return NULL;
2229 if (bs->dirty_bitmap) {
2230 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2231 opaque);
2232 cb = &block_complete_cb;
2233 opaque = blk_cb_data;
2236 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2237 cb, opaque);
2239 if (ret) {
2240 /* Update stats even though technically transfer has not happened. */
2241 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2242 bs->wr_ops ++;
2243 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2244 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2248 return ret;
2252 typedef struct MultiwriteCB {
2253 int error;
2254 int num_requests;
2255 int num_callbacks;
2256 struct {
2257 BlockDriverCompletionFunc *cb;
2258 void *opaque;
2259 QEMUIOVector *free_qiov;
2260 void *free_buf;
2261 } callbacks[];
2262 } MultiwriteCB;
2264 static void multiwrite_user_cb(MultiwriteCB *mcb)
2266 int i;
2268 for (i = 0; i < mcb->num_callbacks; i++) {
2269 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2270 if (mcb->callbacks[i].free_qiov) {
2271 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2273 qemu_free(mcb->callbacks[i].free_qiov);
2274 qemu_vfree(mcb->callbacks[i].free_buf);
2278 static void multiwrite_cb(void *opaque, int ret)
2280 MultiwriteCB *mcb = opaque;
2282 trace_multiwrite_cb(mcb, ret);
2284 if (ret < 0 && !mcb->error) {
2285 mcb->error = ret;
2288 mcb->num_requests--;
2289 if (mcb->num_requests == 0) {
2290 multiwrite_user_cb(mcb);
2291 qemu_free(mcb);
2295 static int multiwrite_req_compare(const void *a, const void *b)
2297 const BlockRequest *req1 = a, *req2 = b;
2300 * Note that we can't simply subtract req2->sector from req1->sector
2301 * here as that could overflow the return value.
2303 if (req1->sector > req2->sector) {
2304 return 1;
2305 } else if (req1->sector < req2->sector) {
2306 return -1;
2307 } else {
2308 return 0;
2313 * Takes a bunch of requests and tries to merge them. Returns the number of
2314 * requests that remain after merging.
2316 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2317 int num_reqs, MultiwriteCB *mcb)
2319 int i, outidx;
2321 // Sort requests by start sector
2322 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2324 // Check if adjacent requests touch the same clusters. If so, combine them,
2325 // filling up gaps with zero sectors.
2326 outidx = 0;
2327 for (i = 1; i < num_reqs; i++) {
2328 int merge = 0;
2329 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2331 // This handles the cases that are valid for all block drivers, namely
2332 // exactly sequential writes and overlapping writes.
2333 if (reqs[i].sector <= oldreq_last) {
2334 merge = 1;
2337 // The block driver may decide that it makes sense to combine requests
2338 // even if there is a gap of some sectors between them. In this case,
2339 // the gap is filled with zeros (therefore only applicable for yet
2340 // unused space in format like qcow2).
2341 if (!merge && bs->drv->bdrv_merge_requests) {
2342 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2345 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2346 merge = 0;
2349 if (merge) {
2350 size_t size;
2351 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2352 qemu_iovec_init(qiov,
2353 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2355 // Add the first request to the merged one. If the requests are
2356 // overlapping, drop the last sectors of the first request.
2357 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2358 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2360 // We might need to add some zeros between the two requests
2361 if (reqs[i].sector > oldreq_last) {
2362 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2363 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2364 memset(buf, 0, zero_bytes);
2365 qemu_iovec_add(qiov, buf, zero_bytes);
2366 mcb->callbacks[i].free_buf = buf;
2369 // Add the second request
2370 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2372 reqs[outidx].nb_sectors = qiov->size >> 9;
2373 reqs[outidx].qiov = qiov;
2375 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2376 } else {
2377 outidx++;
2378 reqs[outidx].sector = reqs[i].sector;
2379 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2380 reqs[outidx].qiov = reqs[i].qiov;
2384 return outidx + 1;
2388 * Submit multiple AIO write requests at once.
2390 * On success, the function returns 0 and all requests in the reqs array have
2391 * been submitted. In error case this function returns -1, and any of the
2392 * requests may or may not be submitted yet. In particular, this means that the
2393 * callback will be called for some of the requests, for others it won't. The
2394 * caller must check the error field of the BlockRequest to wait for the right
2395 * callbacks (if error != 0, no callback will be called).
2397 * The implementation may modify the contents of the reqs array, e.g. to merge
2398 * requests. However, the fields opaque and error are left unmodified as they
2399 * are used to signal failure for a single request to the caller.
2401 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2403 BlockDriverAIOCB *acb;
2404 MultiwriteCB *mcb;
2405 int i;
2407 /* don't submit writes if we don't have a medium */
2408 if (bs->drv == NULL) {
2409 for (i = 0; i < num_reqs; i++) {
2410 reqs[i].error = -ENOMEDIUM;
2412 return -1;
2415 if (num_reqs == 0) {
2416 return 0;
2419 // Create MultiwriteCB structure
2420 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2421 mcb->num_requests = 0;
2422 mcb->num_callbacks = num_reqs;
2424 for (i = 0; i < num_reqs; i++) {
2425 mcb->callbacks[i].cb = reqs[i].cb;
2426 mcb->callbacks[i].opaque = reqs[i].opaque;
2429 // Check for mergable requests
2430 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2432 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2435 * Run the aio requests. As soon as one request can't be submitted
2436 * successfully, fail all requests that are not yet submitted (we must
2437 * return failure for all requests anyway)
2439 * num_requests cannot be set to the right value immediately: If
2440 * bdrv_aio_writev fails for some request, num_requests would be too high
2441 * and therefore multiwrite_cb() would never recognize the multiwrite
2442 * request as completed. We also cannot use the loop variable i to set it
2443 * when the first request fails because the callback may already have been
2444 * called for previously submitted requests. Thus, num_requests must be
2445 * incremented for each request that is submitted.
2447 * The problem that callbacks may be called early also means that we need
2448 * to take care that num_requests doesn't become 0 before all requests are
2449 * submitted - multiwrite_cb() would consider the multiwrite request
2450 * completed. A dummy request that is "completed" by a manual call to
2451 * multiwrite_cb() takes care of this.
2453 mcb->num_requests = 1;
2455 // Run the aio requests
2456 for (i = 0; i < num_reqs; i++) {
2457 mcb->num_requests++;
2458 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2459 reqs[i].nb_sectors, multiwrite_cb, mcb);
2461 if (acb == NULL) {
2462 // We can only fail the whole thing if no request has been
2463 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2464 // complete and report the error in the callback.
2465 if (i == 0) {
2466 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2467 goto fail;
2468 } else {
2469 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2470 multiwrite_cb(mcb, -EIO);
2471 break;
2476 /* Complete the dummy request */
2477 multiwrite_cb(mcb, 0);
2479 return 0;
2481 fail:
2482 for (i = 0; i < mcb->num_callbacks; i++) {
2483 reqs[i].error = -EIO;
2485 qemu_free(mcb);
2486 return -1;
2489 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2490 BlockDriverCompletionFunc *cb, void *opaque)
2492 BlockDriver *drv = bs->drv;
2494 trace_bdrv_aio_flush(bs, opaque);
2496 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2497 return bdrv_aio_noop_em(bs, cb, opaque);
2500 if (!drv)
2501 return NULL;
2502 return drv->bdrv_aio_flush(bs, cb, opaque);
2505 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2507 acb->pool->cancel(acb);
2511 /**************************************************************/
2512 /* async block device emulation */
2514 typedef struct BlockDriverAIOCBSync {
2515 BlockDriverAIOCB common;
2516 QEMUBH *bh;
2517 int ret;
2518 /* vector translation state */
2519 QEMUIOVector *qiov;
2520 uint8_t *bounce;
2521 int is_write;
2522 } BlockDriverAIOCBSync;
2524 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2526 BlockDriverAIOCBSync *acb =
2527 container_of(blockacb, BlockDriverAIOCBSync, common);
2528 qemu_bh_delete(acb->bh);
2529 acb->bh = NULL;
2530 qemu_aio_release(acb);
2533 static AIOPool bdrv_em_aio_pool = {
2534 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2535 .cancel = bdrv_aio_cancel_em,
2538 static void bdrv_aio_bh_cb(void *opaque)
2540 BlockDriverAIOCBSync *acb = opaque;
2542 if (!acb->is_write)
2543 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2544 qemu_vfree(acb->bounce);
2545 acb->common.cb(acb->common.opaque, acb->ret);
2546 qemu_bh_delete(acb->bh);
2547 acb->bh = NULL;
2548 qemu_aio_release(acb);
2551 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2552 int64_t sector_num,
2553 QEMUIOVector *qiov,
2554 int nb_sectors,
2555 BlockDriverCompletionFunc *cb,
2556 void *opaque,
2557 int is_write)
2560 BlockDriverAIOCBSync *acb;
2562 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2563 acb->is_write = is_write;
2564 acb->qiov = qiov;
2565 acb->bounce = qemu_blockalign(bs, qiov->size);
2567 if (!acb->bh)
2568 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2570 if (is_write) {
2571 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2572 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2573 } else {
2574 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2577 qemu_bh_schedule(acb->bh);
2579 return &acb->common;
2582 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2583 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2584 BlockDriverCompletionFunc *cb, void *opaque)
2586 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2589 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2590 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2591 BlockDriverCompletionFunc *cb, void *opaque)
2593 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2596 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2597 BlockDriverCompletionFunc *cb, void *opaque)
2599 BlockDriverAIOCBSync *acb;
2601 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2602 acb->is_write = 1; /* don't bounce in the completion hadler */
2603 acb->qiov = NULL;
2604 acb->bounce = NULL;
2605 acb->ret = 0;
2607 if (!acb->bh)
2608 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2610 bdrv_flush(bs);
2611 qemu_bh_schedule(acb->bh);
2612 return &acb->common;
2615 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2616 BlockDriverCompletionFunc *cb, void *opaque)
2618 BlockDriverAIOCBSync *acb;
2620 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2621 acb->is_write = 1; /* don't bounce in the completion handler */
2622 acb->qiov = NULL;
2623 acb->bounce = NULL;
2624 acb->ret = 0;
2626 if (!acb->bh) {
2627 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2630 qemu_bh_schedule(acb->bh);
2631 return &acb->common;
2634 /**************************************************************/
2635 /* sync block device emulation */
2637 static void bdrv_rw_em_cb(void *opaque, int ret)
2639 *(int *)opaque = ret;
2642 #define NOT_DONE 0x7fffffff
2644 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2645 uint8_t *buf, int nb_sectors)
2647 int async_ret;
2648 BlockDriverAIOCB *acb;
2649 struct iovec iov;
2650 QEMUIOVector qiov;
2652 async_context_push();
2654 async_ret = NOT_DONE;
2655 iov.iov_base = (void *)buf;
2656 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2657 qemu_iovec_init_external(&qiov, &iov, 1);
2658 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2659 bdrv_rw_em_cb, &async_ret);
2660 if (acb == NULL) {
2661 async_ret = -1;
2662 goto fail;
2665 while (async_ret == NOT_DONE) {
2666 qemu_aio_wait();
2670 fail:
2671 async_context_pop();
2672 return async_ret;
2675 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2676 const uint8_t *buf, int nb_sectors)
2678 int async_ret;
2679 BlockDriverAIOCB *acb;
2680 struct iovec iov;
2681 QEMUIOVector qiov;
2683 async_context_push();
2685 async_ret = NOT_DONE;
2686 iov.iov_base = (void *)buf;
2687 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2688 qemu_iovec_init_external(&qiov, &iov, 1);
2689 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2690 bdrv_rw_em_cb, &async_ret);
2691 if (acb == NULL) {
2692 async_ret = -1;
2693 goto fail;
2695 while (async_ret == NOT_DONE) {
2696 qemu_aio_wait();
2699 fail:
2700 async_context_pop();
2701 return async_ret;
2704 void bdrv_init(void)
2706 module_call_init(MODULE_INIT_BLOCK);
2709 void bdrv_init_with_whitelist(void)
2711 use_bdrv_whitelist = 1;
2712 bdrv_init();
2715 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2716 BlockDriverCompletionFunc *cb, void *opaque)
2718 BlockDriverAIOCB *acb;
2720 if (pool->free_aiocb) {
2721 acb = pool->free_aiocb;
2722 pool->free_aiocb = acb->next;
2723 } else {
2724 acb = qemu_mallocz(pool->aiocb_size);
2725 acb->pool = pool;
2727 acb->bs = bs;
2728 acb->cb = cb;
2729 acb->opaque = opaque;
2730 return acb;
2733 void qemu_aio_release(void *p)
2735 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2736 AIOPool *pool = acb->pool;
2737 acb->next = pool->free_aiocb;
2738 pool->free_aiocb = acb;
2741 /**************************************************************/
2742 /* removable device support */
2745 * Return TRUE if the media is present
2747 int bdrv_is_inserted(BlockDriverState *bs)
2749 BlockDriver *drv = bs->drv;
2750 int ret;
2751 if (!drv)
2752 return 0;
2753 if (!drv->bdrv_is_inserted)
2754 return !bs->tray_open;
2755 ret = drv->bdrv_is_inserted(bs);
2756 return ret;
2760 * Return TRUE if the media changed since the last call to this
2761 * function. It is currently only used for floppy disks
2763 int bdrv_media_changed(BlockDriverState *bs)
2765 BlockDriver *drv = bs->drv;
2766 int ret;
2768 if (!drv || !drv->bdrv_media_changed)
2769 ret = -ENOTSUP;
2770 else
2771 ret = drv->bdrv_media_changed(bs);
2772 if (ret == -ENOTSUP)
2773 ret = bs->media_changed;
2774 bs->media_changed = 0;
2775 return ret;
2779 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2781 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2783 BlockDriver *drv = bs->drv;
2784 int ret;
2786 if (bs->locked) {
2787 return -EBUSY;
2790 if (!drv || !drv->bdrv_eject) {
2791 ret = -ENOTSUP;
2792 } else {
2793 ret = drv->bdrv_eject(bs, eject_flag);
2795 if (ret == -ENOTSUP) {
2796 ret = 0;
2798 if (ret >= 0) {
2799 bs->tray_open = eject_flag;
2802 return ret;
2805 int bdrv_is_locked(BlockDriverState *bs)
2807 return bs->locked;
2811 * Lock or unlock the media (if it is locked, the user won't be able
2812 * to eject it manually).
2814 void bdrv_set_locked(BlockDriverState *bs, int locked)
2816 BlockDriver *drv = bs->drv;
2818 trace_bdrv_set_locked(bs, locked);
2820 bs->locked = locked;
2821 if (drv && drv->bdrv_set_locked) {
2822 drv->bdrv_set_locked(bs, locked);
2826 /* needed for generic scsi interface */
2828 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2830 BlockDriver *drv = bs->drv;
2832 if (drv && drv->bdrv_ioctl)
2833 return drv->bdrv_ioctl(bs, req, buf);
2834 return -ENOTSUP;
2837 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2838 unsigned long int req, void *buf,
2839 BlockDriverCompletionFunc *cb, void *opaque)
2841 BlockDriver *drv = bs->drv;
2843 if (drv && drv->bdrv_aio_ioctl)
2844 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2845 return NULL;
2850 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2852 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2855 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2857 int64_t bitmap_size;
2859 bs->dirty_count = 0;
2860 if (enable) {
2861 if (!bs->dirty_bitmap) {
2862 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2863 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2864 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2866 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2868 } else {
2869 if (bs->dirty_bitmap) {
2870 qemu_free(bs->dirty_bitmap);
2871 bs->dirty_bitmap = NULL;
2876 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2878 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2880 if (bs->dirty_bitmap &&
2881 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2882 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2883 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2884 } else {
2885 return 0;
2889 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2890 int nr_sectors)
2892 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2895 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2897 return bs->dirty_count;
2900 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
2902 assert(bs->in_use != in_use);
2903 bs->in_use = in_use;
2906 int bdrv_in_use(BlockDriverState *bs)
2908 return bs->in_use;
2911 int bdrv_img_create(const char *filename, const char *fmt,
2912 const char *base_filename, const char *base_fmt,
2913 char *options, uint64_t img_size, int flags)
2915 QEMUOptionParameter *param = NULL, *create_options = NULL;
2916 QEMUOptionParameter *backing_fmt, *backing_file;
2917 BlockDriverState *bs = NULL;
2918 BlockDriver *drv, *proto_drv;
2919 BlockDriver *backing_drv = NULL;
2920 int ret = 0;
2922 /* Find driver and parse its options */
2923 drv = bdrv_find_format(fmt);
2924 if (!drv) {
2925 error_report("Unknown file format '%s'", fmt);
2926 ret = -EINVAL;
2927 goto out;
2930 proto_drv = bdrv_find_protocol(filename);
2931 if (!proto_drv) {
2932 error_report("Unknown protocol '%s'", filename);
2933 ret = -EINVAL;
2934 goto out;
2937 create_options = append_option_parameters(create_options,
2938 drv->create_options);
2939 create_options = append_option_parameters(create_options,
2940 proto_drv->create_options);
2942 /* Create parameter list with default values */
2943 param = parse_option_parameters("", create_options, param);
2945 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
2947 /* Parse -o options */
2948 if (options) {
2949 param = parse_option_parameters(options, create_options, param);
2950 if (param == NULL) {
2951 error_report("Invalid options for file format '%s'.", fmt);
2952 ret = -EINVAL;
2953 goto out;
2957 if (base_filename) {
2958 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
2959 base_filename)) {
2960 error_report("Backing file not supported for file format '%s'",
2961 fmt);
2962 ret = -EINVAL;
2963 goto out;
2967 if (base_fmt) {
2968 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
2969 error_report("Backing file format not supported for file "
2970 "format '%s'", fmt);
2971 ret = -EINVAL;
2972 goto out;
2976 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
2977 if (backing_file && backing_file->value.s) {
2978 if (!strcmp(filename, backing_file->value.s)) {
2979 error_report("Error: Trying to create an image with the "
2980 "same filename as the backing file");
2981 ret = -EINVAL;
2982 goto out;
2986 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
2987 if (backing_fmt && backing_fmt->value.s) {
2988 backing_drv = bdrv_find_format(backing_fmt->value.s);
2989 if (!backing_drv) {
2990 error_report("Unknown backing file format '%s'",
2991 backing_fmt->value.s);
2992 ret = -EINVAL;
2993 goto out;
2997 // The size for the image must always be specified, with one exception:
2998 // If we are using a backing file, we can obtain the size from there
2999 if (get_option_parameter(param, BLOCK_OPT_SIZE)->value.n == -1) {
3000 if (backing_file && backing_file->value.s) {
3001 uint64_t size;
3002 char buf[32];
3004 bs = bdrv_new("");
3006 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
3007 if (ret < 0) {
3008 error_report("Could not open '%s'", backing_file->value.s);
3009 goto out;
3011 bdrv_get_geometry(bs, &size);
3012 size *= 512;
3014 snprintf(buf, sizeof(buf), "%" PRId64, size);
3015 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3016 } else {
3017 error_report("Image creation needs a size parameter");
3018 ret = -EINVAL;
3019 goto out;
3023 printf("Formatting '%s', fmt=%s ", filename, fmt);
3024 print_option_parameters(param);
3025 puts("");
3027 ret = bdrv_create(drv, filename, param);
3029 if (ret < 0) {
3030 if (ret == -ENOTSUP) {
3031 error_report("Formatting or formatting option not supported for "
3032 "file format '%s'", fmt);
3033 } else if (ret == -EFBIG) {
3034 error_report("The image size is too large for file format '%s'",
3035 fmt);
3036 } else {
3037 error_report("%s: error while creating %s: %s", filename, fmt,
3038 strerror(-ret));
3042 out:
3043 free_option_parameters(create_options);
3044 free_option_parameters(param);
3046 if (bs) {
3047 bdrv_delete(bs);
3050 return ret;