Add support for finding libpng via pkg-config.
[qemu.git] / block.c
blob9549b9eff9f3034fdfae86e239851db459b9c0da
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);
442 if (flags & BDRV_O_CACHE_WB)
443 bs->enable_write_cache = 1;
446 * Clear flags that are internal to the block layer before opening the
447 * image.
449 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
452 * Snapshots should be writable.
454 if (bs->is_temporary) {
455 open_flags |= BDRV_O_RDWR;
458 /* Open the image, either directly or using a protocol */
459 if (drv->bdrv_file_open) {
460 ret = drv->bdrv_file_open(bs, filename, open_flags);
461 } else {
462 ret = bdrv_file_open(&bs->file, filename, open_flags);
463 if (ret >= 0) {
464 ret = drv->bdrv_open(bs, open_flags);
468 if (ret < 0) {
469 goto free_and_fail;
472 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
474 ret = refresh_total_sectors(bs, bs->total_sectors);
475 if (ret < 0) {
476 goto free_and_fail;
479 #ifndef _WIN32
480 if (bs->is_temporary) {
481 unlink(filename);
483 #endif
484 return 0;
486 free_and_fail:
487 if (bs->file) {
488 bdrv_delete(bs->file);
489 bs->file = NULL;
491 qemu_free(bs->opaque);
492 bs->opaque = NULL;
493 bs->drv = NULL;
494 return ret;
498 * Opens a file using a protocol (file, host_device, nbd, ...)
500 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
502 BlockDriverState *bs;
503 BlockDriver *drv;
504 int ret;
506 drv = bdrv_find_protocol(filename);
507 if (!drv) {
508 return -ENOENT;
511 bs = bdrv_new("");
512 ret = bdrv_open_common(bs, filename, flags, drv);
513 if (ret < 0) {
514 bdrv_delete(bs);
515 return ret;
517 bs->growable = 1;
518 *pbs = bs;
519 return 0;
523 * Opens a disk image (raw, qcow2, vmdk, ...)
525 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
526 BlockDriver *drv)
528 int ret;
530 if (flags & BDRV_O_SNAPSHOT) {
531 BlockDriverState *bs1;
532 int64_t total_size;
533 int is_protocol = 0;
534 BlockDriver *bdrv_qcow2;
535 QEMUOptionParameter *options;
536 char tmp_filename[PATH_MAX];
537 char backing_filename[PATH_MAX];
539 /* if snapshot, we create a temporary backing file and open it
540 instead of opening 'filename' directly */
542 /* if there is a backing file, use it */
543 bs1 = bdrv_new("");
544 ret = bdrv_open(bs1, filename, 0, drv);
545 if (ret < 0) {
546 bdrv_delete(bs1);
547 return ret;
549 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
551 if (bs1->drv && bs1->drv->protocol_name)
552 is_protocol = 1;
554 bdrv_delete(bs1);
556 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
558 /* Real path is meaningless for protocols */
559 if (is_protocol)
560 snprintf(backing_filename, sizeof(backing_filename),
561 "%s", filename);
562 else if (!realpath(filename, backing_filename))
563 return -errno;
565 bdrv_qcow2 = bdrv_find_format("qcow2");
566 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
568 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
569 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
570 if (drv) {
571 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
572 drv->format_name);
575 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
576 free_option_parameters(options);
577 if (ret < 0) {
578 return ret;
581 filename = tmp_filename;
582 drv = bdrv_qcow2;
583 bs->is_temporary = 1;
586 /* Find the right image format driver */
587 if (!drv) {
588 ret = find_image_format(filename, &drv);
591 if (!drv) {
592 goto unlink_and_fail;
595 /* Open the image */
596 ret = bdrv_open_common(bs, filename, flags, drv);
597 if (ret < 0) {
598 goto unlink_and_fail;
601 /* If there is a backing file, use it */
602 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
603 char backing_filename[PATH_MAX];
604 int back_flags;
605 BlockDriver *back_drv = NULL;
607 bs->backing_hd = bdrv_new("");
609 if (path_has_protocol(bs->backing_file)) {
610 pstrcpy(backing_filename, sizeof(backing_filename),
611 bs->backing_file);
612 } else {
613 path_combine(backing_filename, sizeof(backing_filename),
614 filename, bs->backing_file);
617 if (bs->backing_format[0] != '\0') {
618 back_drv = bdrv_find_format(bs->backing_format);
621 /* backing files always opened read-only */
622 back_flags =
623 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
625 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
626 if (ret < 0) {
627 bdrv_close(bs);
628 return ret;
630 if (bs->is_temporary) {
631 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
632 } else {
633 /* base image inherits from "parent" */
634 bs->backing_hd->keep_read_only = bs->keep_read_only;
638 if (!bdrv_key_required(bs)) {
639 /* call the change callback */
640 bs->media_changed = 1;
641 if (bs->change_cb)
642 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
645 return 0;
647 unlink_and_fail:
648 if (bs->is_temporary) {
649 unlink(filename);
651 return ret;
654 void bdrv_close(BlockDriverState *bs)
656 if (bs->drv) {
657 if (bs == bs_snapshots) {
658 bs_snapshots = NULL;
660 if (bs->backing_hd) {
661 bdrv_delete(bs->backing_hd);
662 bs->backing_hd = NULL;
664 bs->drv->bdrv_close(bs);
665 qemu_free(bs->opaque);
666 #ifdef _WIN32
667 if (bs->is_temporary) {
668 unlink(bs->filename);
670 #endif
671 bs->opaque = NULL;
672 bs->drv = NULL;
674 if (bs->file != NULL) {
675 bdrv_close(bs->file);
678 /* call the change callback */
679 bs->media_changed = 1;
680 if (bs->change_cb)
681 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
685 void bdrv_close_all(void)
687 BlockDriverState *bs;
689 QTAILQ_FOREACH(bs, &bdrv_states, list) {
690 bdrv_close(bs);
694 /* make a BlockDriverState anonymous by removing from bdrv_state list.
695 Also, NULL terminate the device_name to prevent double remove */
696 void bdrv_make_anon(BlockDriverState *bs)
698 if (bs->device_name[0] != '\0') {
699 QTAILQ_REMOVE(&bdrv_states, bs, list);
701 bs->device_name[0] = '\0';
704 void bdrv_delete(BlockDriverState *bs)
706 assert(!bs->peer);
708 /* remove from list, if necessary */
709 bdrv_make_anon(bs);
711 bdrv_close(bs);
712 if (bs->file != NULL) {
713 bdrv_delete(bs->file);
716 assert(bs != bs_snapshots);
717 qemu_free(bs);
720 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
722 if (bs->peer) {
723 return -EBUSY;
725 bs->peer = qdev;
726 return 0;
729 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
731 assert(bs->peer == qdev);
732 bs->peer = NULL;
735 DeviceState *bdrv_get_attached(BlockDriverState *bs)
737 return bs->peer;
741 * Run consistency checks on an image
743 * Returns 0 if the check could be completed (it doesn't mean that the image is
744 * free of errors) or -errno when an internal error occurred. The results of the
745 * check are stored in res.
747 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
749 if (bs->drv->bdrv_check == NULL) {
750 return -ENOTSUP;
753 memset(res, 0, sizeof(*res));
754 return bs->drv->bdrv_check(bs, res);
757 #define COMMIT_BUF_SECTORS 2048
759 /* commit COW file into the raw image */
760 int bdrv_commit(BlockDriverState *bs)
762 BlockDriver *drv = bs->drv;
763 BlockDriver *backing_drv;
764 int64_t sector, total_sectors;
765 int n, ro, open_flags;
766 int ret = 0, rw_ret = 0;
767 uint8_t *buf;
768 char filename[1024];
769 BlockDriverState *bs_rw, *bs_ro;
771 if (!drv)
772 return -ENOMEDIUM;
774 if (!bs->backing_hd) {
775 return -ENOTSUP;
778 if (bs->backing_hd->keep_read_only) {
779 return -EACCES;
782 backing_drv = bs->backing_hd->drv;
783 ro = bs->backing_hd->read_only;
784 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
785 open_flags = bs->backing_hd->open_flags;
787 if (ro) {
788 /* re-open as RW */
789 bdrv_delete(bs->backing_hd);
790 bs->backing_hd = NULL;
791 bs_rw = bdrv_new("");
792 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
793 backing_drv);
794 if (rw_ret < 0) {
795 bdrv_delete(bs_rw);
796 /* try to re-open read-only */
797 bs_ro = bdrv_new("");
798 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
799 backing_drv);
800 if (ret < 0) {
801 bdrv_delete(bs_ro);
802 /* drive not functional anymore */
803 bs->drv = NULL;
804 return ret;
806 bs->backing_hd = bs_ro;
807 return rw_ret;
809 bs->backing_hd = bs_rw;
812 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
813 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
815 for (sector = 0; sector < total_sectors; sector += n) {
816 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
818 if (bdrv_read(bs, sector, buf, n) != 0) {
819 ret = -EIO;
820 goto ro_cleanup;
823 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
824 ret = -EIO;
825 goto ro_cleanup;
830 if (drv->bdrv_make_empty) {
831 ret = drv->bdrv_make_empty(bs);
832 bdrv_flush(bs);
836 * Make sure all data we wrote to the backing device is actually
837 * stable on disk.
839 if (bs->backing_hd)
840 bdrv_flush(bs->backing_hd);
842 ro_cleanup:
843 qemu_free(buf);
845 if (ro) {
846 /* re-open as RO */
847 bdrv_delete(bs->backing_hd);
848 bs->backing_hd = NULL;
849 bs_ro = bdrv_new("");
850 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
851 backing_drv);
852 if (ret < 0) {
853 bdrv_delete(bs_ro);
854 /* drive not functional anymore */
855 bs->drv = NULL;
856 return ret;
858 bs->backing_hd = bs_ro;
859 bs->backing_hd->keep_read_only = 0;
862 return ret;
865 void bdrv_commit_all(void)
867 BlockDriverState *bs;
869 QTAILQ_FOREACH(bs, &bdrv_states, list) {
870 bdrv_commit(bs);
875 * Return values:
876 * 0 - success
877 * -EINVAL - backing format specified, but no file
878 * -ENOSPC - can't update the backing file because no space is left in the
879 * image file header
880 * -ENOTSUP - format driver doesn't support changing the backing file
882 int bdrv_change_backing_file(BlockDriverState *bs,
883 const char *backing_file, const char *backing_fmt)
885 BlockDriver *drv = bs->drv;
887 if (drv->bdrv_change_backing_file != NULL) {
888 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
889 } else {
890 return -ENOTSUP;
894 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
895 size_t size)
897 int64_t len;
899 if (!bdrv_is_inserted(bs))
900 return -ENOMEDIUM;
902 if (bs->growable)
903 return 0;
905 len = bdrv_getlength(bs);
907 if (offset < 0)
908 return -EIO;
910 if ((offset > len) || (len - offset < size))
911 return -EIO;
913 return 0;
916 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
917 int nb_sectors)
919 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
920 nb_sectors * BDRV_SECTOR_SIZE);
923 /* return < 0 if error. See bdrv_write() for the return codes */
924 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
925 uint8_t *buf, int nb_sectors)
927 BlockDriver *drv = bs->drv;
929 if (!drv)
930 return -ENOMEDIUM;
931 if (bdrv_check_request(bs, sector_num, nb_sectors))
932 return -EIO;
934 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
937 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
938 int nb_sectors, int dirty)
940 int64_t start, end;
941 unsigned long val, idx, bit;
943 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
944 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
946 for (; start <= end; start++) {
947 idx = start / (sizeof(unsigned long) * 8);
948 bit = start % (sizeof(unsigned long) * 8);
949 val = bs->dirty_bitmap[idx];
950 if (dirty) {
951 if (!(val & (1UL << bit))) {
952 bs->dirty_count++;
953 val |= 1UL << bit;
955 } else {
956 if (val & (1UL << bit)) {
957 bs->dirty_count--;
958 val &= ~(1UL << bit);
961 bs->dirty_bitmap[idx] = val;
965 /* Return < 0 if error. Important errors are:
966 -EIO generic I/O error (may happen for all errors)
967 -ENOMEDIUM No media inserted.
968 -EINVAL Invalid sector number or nb_sectors
969 -EACCES Trying to write a read-only device
971 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
972 const uint8_t *buf, int nb_sectors)
974 BlockDriver *drv = bs->drv;
975 if (!bs->drv)
976 return -ENOMEDIUM;
977 if (bs->read_only)
978 return -EACCES;
979 if (bdrv_check_request(bs, sector_num, nb_sectors))
980 return -EIO;
982 if (bs->dirty_bitmap) {
983 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
986 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
987 bs->wr_highest_sector = sector_num + nb_sectors - 1;
990 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
993 int bdrv_pread(BlockDriverState *bs, int64_t offset,
994 void *buf, int count1)
996 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
997 int len, nb_sectors, count;
998 int64_t sector_num;
999 int ret;
1001 count = count1;
1002 /* first read to align to sector start */
1003 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1004 if (len > count)
1005 len = count;
1006 sector_num = offset >> BDRV_SECTOR_BITS;
1007 if (len > 0) {
1008 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1009 return ret;
1010 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1011 count -= len;
1012 if (count == 0)
1013 return count1;
1014 sector_num++;
1015 buf += len;
1018 /* read the sectors "in place" */
1019 nb_sectors = count >> BDRV_SECTOR_BITS;
1020 if (nb_sectors > 0) {
1021 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1022 return ret;
1023 sector_num += nb_sectors;
1024 len = nb_sectors << BDRV_SECTOR_BITS;
1025 buf += len;
1026 count -= len;
1029 /* add data from the last sector */
1030 if (count > 0) {
1031 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1032 return ret;
1033 memcpy(buf, tmp_buf, count);
1035 return count1;
1038 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1039 const void *buf, int count1)
1041 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1042 int len, nb_sectors, count;
1043 int64_t sector_num;
1044 int ret;
1046 count = count1;
1047 /* first write to align to sector start */
1048 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1049 if (len > count)
1050 len = count;
1051 sector_num = offset >> BDRV_SECTOR_BITS;
1052 if (len > 0) {
1053 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1054 return ret;
1055 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1056 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1057 return ret;
1058 count -= len;
1059 if (count == 0)
1060 return count1;
1061 sector_num++;
1062 buf += len;
1065 /* write the sectors "in place" */
1066 nb_sectors = count >> BDRV_SECTOR_BITS;
1067 if (nb_sectors > 0) {
1068 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1069 return ret;
1070 sector_num += nb_sectors;
1071 len = nb_sectors << BDRV_SECTOR_BITS;
1072 buf += len;
1073 count -= len;
1076 /* add data from the last sector */
1077 if (count > 0) {
1078 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1079 return ret;
1080 memcpy(tmp_buf, buf, count);
1081 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1082 return ret;
1084 return count1;
1088 * Writes to the file and ensures that no writes are reordered across this
1089 * request (acts as a barrier)
1091 * Returns 0 on success, -errno in error cases.
1093 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1094 const void *buf, int count)
1096 int ret;
1098 ret = bdrv_pwrite(bs, offset, buf, count);
1099 if (ret < 0) {
1100 return ret;
1103 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1104 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1105 bdrv_flush(bs);
1108 return 0;
1112 * Writes to the file and ensures that no writes are reordered across this
1113 * request (acts as a barrier)
1115 * Returns 0 on success, -errno in error cases.
1117 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1118 const uint8_t *buf, int nb_sectors)
1120 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1121 buf, BDRV_SECTOR_SIZE * nb_sectors);
1125 * Truncate file to 'offset' bytes (needed only for file protocols)
1127 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1129 BlockDriver *drv = bs->drv;
1130 int ret;
1131 if (!drv)
1132 return -ENOMEDIUM;
1133 if (!drv->bdrv_truncate)
1134 return -ENOTSUP;
1135 if (bs->read_only)
1136 return -EACCES;
1137 if (bdrv_in_use(bs))
1138 return -EBUSY;
1139 ret = drv->bdrv_truncate(bs, offset);
1140 if (ret == 0) {
1141 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1142 if (bs->change_cb) {
1143 bs->change_cb(bs->change_opaque, CHANGE_SIZE);
1146 return ret;
1150 * Length of a allocated file in bytes. Sparse files are counted by actual
1151 * allocated space. Return < 0 if error or unknown.
1153 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1155 BlockDriver *drv = bs->drv;
1156 if (!drv) {
1157 return -ENOMEDIUM;
1159 if (drv->bdrv_get_allocated_file_size) {
1160 return drv->bdrv_get_allocated_file_size(bs);
1162 if (bs->file) {
1163 return bdrv_get_allocated_file_size(bs->file);
1165 return -ENOTSUP;
1169 * Length of a file in bytes. Return < 0 if error or unknown.
1171 int64_t bdrv_getlength(BlockDriverState *bs)
1173 BlockDriver *drv = bs->drv;
1174 if (!drv)
1175 return -ENOMEDIUM;
1177 if (bs->growable || bs->removable) {
1178 if (drv->bdrv_getlength) {
1179 return drv->bdrv_getlength(bs);
1182 return bs->total_sectors * BDRV_SECTOR_SIZE;
1185 /* return 0 as number of sectors if no device present or error */
1186 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1188 int64_t length;
1189 length = bdrv_getlength(bs);
1190 if (length < 0)
1191 length = 0;
1192 else
1193 length = length >> BDRV_SECTOR_BITS;
1194 *nb_sectors_ptr = length;
1197 struct partition {
1198 uint8_t boot_ind; /* 0x80 - active */
1199 uint8_t head; /* starting head */
1200 uint8_t sector; /* starting sector */
1201 uint8_t cyl; /* starting cylinder */
1202 uint8_t sys_ind; /* What partition type */
1203 uint8_t end_head; /* end head */
1204 uint8_t end_sector; /* end sector */
1205 uint8_t end_cyl; /* end cylinder */
1206 uint32_t start_sect; /* starting sector counting from 0 */
1207 uint32_t nr_sects; /* nr of sectors in partition */
1208 } __attribute__((packed));
1210 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1211 static int guess_disk_lchs(BlockDriverState *bs,
1212 int *pcylinders, int *pheads, int *psectors)
1214 uint8_t buf[BDRV_SECTOR_SIZE];
1215 int ret, i, heads, sectors, cylinders;
1216 struct partition *p;
1217 uint32_t nr_sects;
1218 uint64_t nb_sectors;
1220 bdrv_get_geometry(bs, &nb_sectors);
1222 ret = bdrv_read(bs, 0, buf, 1);
1223 if (ret < 0)
1224 return -1;
1225 /* test msdos magic */
1226 if (buf[510] != 0x55 || buf[511] != 0xaa)
1227 return -1;
1228 for(i = 0; i < 4; i++) {
1229 p = ((struct partition *)(buf + 0x1be)) + i;
1230 nr_sects = le32_to_cpu(p->nr_sects);
1231 if (nr_sects && p->end_head) {
1232 /* We make the assumption that the partition terminates on
1233 a cylinder boundary */
1234 heads = p->end_head + 1;
1235 sectors = p->end_sector & 63;
1236 if (sectors == 0)
1237 continue;
1238 cylinders = nb_sectors / (heads * sectors);
1239 if (cylinders < 1 || cylinders > 16383)
1240 continue;
1241 *pheads = heads;
1242 *psectors = sectors;
1243 *pcylinders = cylinders;
1244 #if 0
1245 printf("guessed geometry: LCHS=%d %d %d\n",
1246 cylinders, heads, sectors);
1247 #endif
1248 return 0;
1251 return -1;
1254 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1256 int translation, lba_detected = 0;
1257 int cylinders, heads, secs;
1258 uint64_t nb_sectors;
1260 /* if a geometry hint is available, use it */
1261 bdrv_get_geometry(bs, &nb_sectors);
1262 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1263 translation = bdrv_get_translation_hint(bs);
1264 if (cylinders != 0) {
1265 *pcyls = cylinders;
1266 *pheads = heads;
1267 *psecs = secs;
1268 } else {
1269 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1270 if (heads > 16) {
1271 /* if heads > 16, it means that a BIOS LBA
1272 translation was active, so the default
1273 hardware geometry is OK */
1274 lba_detected = 1;
1275 goto default_geometry;
1276 } else {
1277 *pcyls = cylinders;
1278 *pheads = heads;
1279 *psecs = secs;
1280 /* disable any translation to be in sync with
1281 the logical geometry */
1282 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1283 bdrv_set_translation_hint(bs,
1284 BIOS_ATA_TRANSLATION_NONE);
1287 } else {
1288 default_geometry:
1289 /* if no geometry, use a standard physical disk geometry */
1290 cylinders = nb_sectors / (16 * 63);
1292 if (cylinders > 16383)
1293 cylinders = 16383;
1294 else if (cylinders < 2)
1295 cylinders = 2;
1296 *pcyls = cylinders;
1297 *pheads = 16;
1298 *psecs = 63;
1299 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1300 if ((*pcyls * *pheads) <= 131072) {
1301 bdrv_set_translation_hint(bs,
1302 BIOS_ATA_TRANSLATION_LARGE);
1303 } else {
1304 bdrv_set_translation_hint(bs,
1305 BIOS_ATA_TRANSLATION_LBA);
1309 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1313 void bdrv_set_geometry_hint(BlockDriverState *bs,
1314 int cyls, int heads, int secs)
1316 bs->cyls = cyls;
1317 bs->heads = heads;
1318 bs->secs = secs;
1321 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1323 bs->translation = translation;
1326 void bdrv_get_geometry_hint(BlockDriverState *bs,
1327 int *pcyls, int *pheads, int *psecs)
1329 *pcyls = bs->cyls;
1330 *pheads = bs->heads;
1331 *psecs = bs->secs;
1334 /* Recognize floppy formats */
1335 typedef struct FDFormat {
1336 FDriveType drive;
1337 uint8_t last_sect;
1338 uint8_t max_track;
1339 uint8_t max_head;
1340 } FDFormat;
1342 static const FDFormat fd_formats[] = {
1343 /* First entry is default format */
1344 /* 1.44 MB 3"1/2 floppy disks */
1345 { FDRIVE_DRV_144, 18, 80, 1, },
1346 { FDRIVE_DRV_144, 20, 80, 1, },
1347 { FDRIVE_DRV_144, 21, 80, 1, },
1348 { FDRIVE_DRV_144, 21, 82, 1, },
1349 { FDRIVE_DRV_144, 21, 83, 1, },
1350 { FDRIVE_DRV_144, 22, 80, 1, },
1351 { FDRIVE_DRV_144, 23, 80, 1, },
1352 { FDRIVE_DRV_144, 24, 80, 1, },
1353 /* 2.88 MB 3"1/2 floppy disks */
1354 { FDRIVE_DRV_288, 36, 80, 1, },
1355 { FDRIVE_DRV_288, 39, 80, 1, },
1356 { FDRIVE_DRV_288, 40, 80, 1, },
1357 { FDRIVE_DRV_288, 44, 80, 1, },
1358 { FDRIVE_DRV_288, 48, 80, 1, },
1359 /* 720 kB 3"1/2 floppy disks */
1360 { FDRIVE_DRV_144, 9, 80, 1, },
1361 { FDRIVE_DRV_144, 10, 80, 1, },
1362 { FDRIVE_DRV_144, 10, 82, 1, },
1363 { FDRIVE_DRV_144, 10, 83, 1, },
1364 { FDRIVE_DRV_144, 13, 80, 1, },
1365 { FDRIVE_DRV_144, 14, 80, 1, },
1366 /* 1.2 MB 5"1/4 floppy disks */
1367 { FDRIVE_DRV_120, 15, 80, 1, },
1368 { FDRIVE_DRV_120, 18, 80, 1, },
1369 { FDRIVE_DRV_120, 18, 82, 1, },
1370 { FDRIVE_DRV_120, 18, 83, 1, },
1371 { FDRIVE_DRV_120, 20, 80, 1, },
1372 /* 720 kB 5"1/4 floppy disks */
1373 { FDRIVE_DRV_120, 9, 80, 1, },
1374 { FDRIVE_DRV_120, 11, 80, 1, },
1375 /* 360 kB 5"1/4 floppy disks */
1376 { FDRIVE_DRV_120, 9, 40, 1, },
1377 { FDRIVE_DRV_120, 9, 40, 0, },
1378 { FDRIVE_DRV_120, 10, 41, 1, },
1379 { FDRIVE_DRV_120, 10, 42, 1, },
1380 /* 320 kB 5"1/4 floppy disks */
1381 { FDRIVE_DRV_120, 8, 40, 1, },
1382 { FDRIVE_DRV_120, 8, 40, 0, },
1383 /* 360 kB must match 5"1/4 better than 3"1/2... */
1384 { FDRIVE_DRV_144, 9, 80, 0, },
1385 /* end */
1386 { FDRIVE_DRV_NONE, -1, -1, 0, },
1389 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
1390 int *max_track, int *last_sect,
1391 FDriveType drive_in, FDriveType *drive)
1393 const FDFormat *parse;
1394 uint64_t nb_sectors, size;
1395 int i, first_match, match;
1397 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
1398 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
1399 /* User defined disk */
1400 } else {
1401 bdrv_get_geometry(bs, &nb_sectors);
1402 match = -1;
1403 first_match = -1;
1404 for (i = 0; ; i++) {
1405 parse = &fd_formats[i];
1406 if (parse->drive == FDRIVE_DRV_NONE) {
1407 break;
1409 if (drive_in == parse->drive ||
1410 drive_in == FDRIVE_DRV_NONE) {
1411 size = (parse->max_head + 1) * parse->max_track *
1412 parse->last_sect;
1413 if (nb_sectors == size) {
1414 match = i;
1415 break;
1417 if (first_match == -1) {
1418 first_match = i;
1422 if (match == -1) {
1423 if (first_match == -1) {
1424 match = 1;
1425 } else {
1426 match = first_match;
1428 parse = &fd_formats[match];
1430 *nb_heads = parse->max_head + 1;
1431 *max_track = parse->max_track;
1432 *last_sect = parse->last_sect;
1433 *drive = parse->drive;
1437 int bdrv_get_translation_hint(BlockDriverState *bs)
1439 return bs->translation;
1442 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1443 BlockErrorAction on_write_error)
1445 bs->on_read_error = on_read_error;
1446 bs->on_write_error = on_write_error;
1449 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1451 return is_read ? bs->on_read_error : bs->on_write_error;
1454 void bdrv_set_removable(BlockDriverState *bs, int removable)
1456 bs->removable = removable;
1457 if (removable && bs == bs_snapshots) {
1458 bs_snapshots = NULL;
1462 int bdrv_is_removable(BlockDriverState *bs)
1464 return bs->removable;
1467 int bdrv_is_read_only(BlockDriverState *bs)
1469 return bs->read_only;
1472 int bdrv_is_sg(BlockDriverState *bs)
1474 return bs->sg;
1477 int bdrv_enable_write_cache(BlockDriverState *bs)
1479 return bs->enable_write_cache;
1482 /* XXX: no longer used */
1483 void bdrv_set_change_cb(BlockDriverState *bs,
1484 void (*change_cb)(void *opaque, int reason),
1485 void *opaque)
1487 bs->change_cb = change_cb;
1488 bs->change_opaque = opaque;
1491 int bdrv_is_encrypted(BlockDriverState *bs)
1493 if (bs->backing_hd && bs->backing_hd->encrypted)
1494 return 1;
1495 return bs->encrypted;
1498 int bdrv_key_required(BlockDriverState *bs)
1500 BlockDriverState *backing_hd = bs->backing_hd;
1502 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1503 return 1;
1504 return (bs->encrypted && !bs->valid_key);
1507 int bdrv_set_key(BlockDriverState *bs, const char *key)
1509 int ret;
1510 if (bs->backing_hd && bs->backing_hd->encrypted) {
1511 ret = bdrv_set_key(bs->backing_hd, key);
1512 if (ret < 0)
1513 return ret;
1514 if (!bs->encrypted)
1515 return 0;
1517 if (!bs->encrypted) {
1518 return -EINVAL;
1519 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1520 return -ENOMEDIUM;
1522 ret = bs->drv->bdrv_set_key(bs, key);
1523 if (ret < 0) {
1524 bs->valid_key = 0;
1525 } else if (!bs->valid_key) {
1526 bs->valid_key = 1;
1527 /* call the change callback now, we skipped it on open */
1528 bs->media_changed = 1;
1529 if (bs->change_cb)
1530 bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
1532 return ret;
1535 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1537 if (!bs->drv) {
1538 buf[0] = '\0';
1539 } else {
1540 pstrcpy(buf, buf_size, bs->drv->format_name);
1544 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1545 void *opaque)
1547 BlockDriver *drv;
1549 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1550 it(opaque, drv->format_name);
1554 BlockDriverState *bdrv_find(const char *name)
1556 BlockDriverState *bs;
1558 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1559 if (!strcmp(name, bs->device_name)) {
1560 return bs;
1563 return NULL;
1566 BlockDriverState *bdrv_next(BlockDriverState *bs)
1568 if (!bs) {
1569 return QTAILQ_FIRST(&bdrv_states);
1571 return QTAILQ_NEXT(bs, list);
1574 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1576 BlockDriverState *bs;
1578 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1579 it(opaque, bs);
1583 const char *bdrv_get_device_name(BlockDriverState *bs)
1585 return bs->device_name;
1588 int bdrv_flush(BlockDriverState *bs)
1590 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1591 return 0;
1594 if (bs->drv && bs->drv->bdrv_flush) {
1595 return bs->drv->bdrv_flush(bs);
1599 * Some block drivers always operate in either writethrough or unsafe mode
1600 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1601 * the server works (because the behaviour is hardcoded or depends on
1602 * server-side configuration), so we can't ensure that everything is safe
1603 * on disk. Returning an error doesn't work because that would break guests
1604 * even if the server operates in writethrough mode.
1606 * Let's hope the user knows what he's doing.
1608 return 0;
1611 void bdrv_flush_all(void)
1613 BlockDriverState *bs;
1615 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1616 if (bs->drv && !bdrv_is_read_only(bs) &&
1617 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1618 bdrv_flush(bs);
1623 int bdrv_has_zero_init(BlockDriverState *bs)
1625 assert(bs->drv);
1627 if (bs->drv->bdrv_has_zero_init) {
1628 return bs->drv->bdrv_has_zero_init(bs);
1631 return 1;
1634 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1636 if (!bs->drv) {
1637 return -ENOMEDIUM;
1639 if (!bs->drv->bdrv_discard) {
1640 return 0;
1642 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1646 * Returns true iff the specified sector is present in the disk image. Drivers
1647 * not implementing the functionality are assumed to not support backing files,
1648 * hence all their sectors are reported as allocated.
1650 * 'pnum' is set to the number of sectors (including and immediately following
1651 * the specified sector) that are known to be in the same
1652 * allocated/unallocated state.
1654 * 'nb_sectors' is the max value 'pnum' should be set to.
1656 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1657 int *pnum)
1659 int64_t n;
1660 if (!bs->drv->bdrv_is_allocated) {
1661 if (sector_num >= bs->total_sectors) {
1662 *pnum = 0;
1663 return 0;
1665 n = bs->total_sectors - sector_num;
1666 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1667 return 1;
1669 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1672 void bdrv_mon_event(const BlockDriverState *bdrv,
1673 BlockMonEventAction action, int is_read)
1675 QObject *data;
1676 const char *action_str;
1678 switch (action) {
1679 case BDRV_ACTION_REPORT:
1680 action_str = "report";
1681 break;
1682 case BDRV_ACTION_IGNORE:
1683 action_str = "ignore";
1684 break;
1685 case BDRV_ACTION_STOP:
1686 action_str = "stop";
1687 break;
1688 default:
1689 abort();
1692 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1693 bdrv->device_name,
1694 action_str,
1695 is_read ? "read" : "write");
1696 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1698 qobject_decref(data);
1701 static void bdrv_print_dict(QObject *obj, void *opaque)
1703 QDict *bs_dict;
1704 Monitor *mon = opaque;
1706 bs_dict = qobject_to_qdict(obj);
1708 monitor_printf(mon, "%s: removable=%d",
1709 qdict_get_str(bs_dict, "device"),
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;
1751 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': 'unknown', "
1752 "'removable': %i, 'locked': %i }",
1753 bs->device_name, bs->removable,
1754 bs->locked);
1756 if (bs->drv) {
1757 QObject *obj;
1758 QDict *bs_dict = qobject_to_qdict(bs_obj);
1760 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1761 "'encrypted': %i }",
1762 bs->filename, bs->read_only,
1763 bs->drv->format_name,
1764 bdrv_is_encrypted(bs));
1765 if (bs->backing_file[0] != '\0') {
1766 QDict *qdict = qobject_to_qdict(obj);
1767 qdict_put(qdict, "backing_file",
1768 qstring_from_str(bs->backing_file));
1771 qdict_put_obj(bs_dict, "inserted", obj);
1773 qlist_append_obj(bs_list, bs_obj);
1776 *ret_data = QOBJECT(bs_list);
1779 static void bdrv_stats_iter(QObject *data, void *opaque)
1781 QDict *qdict;
1782 Monitor *mon = opaque;
1784 qdict = qobject_to_qdict(data);
1785 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1787 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1788 monitor_printf(mon, " rd_bytes=%" PRId64
1789 " wr_bytes=%" PRId64
1790 " rd_operations=%" PRId64
1791 " wr_operations=%" PRId64
1792 "\n",
1793 qdict_get_int(qdict, "rd_bytes"),
1794 qdict_get_int(qdict, "wr_bytes"),
1795 qdict_get_int(qdict, "rd_operations"),
1796 qdict_get_int(qdict, "wr_operations"));
1799 void bdrv_stats_print(Monitor *mon, const QObject *data)
1801 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1804 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1806 QObject *res;
1807 QDict *dict;
1809 res = qobject_from_jsonf("{ 'stats': {"
1810 "'rd_bytes': %" PRId64 ","
1811 "'wr_bytes': %" PRId64 ","
1812 "'rd_operations': %" PRId64 ","
1813 "'wr_operations': %" PRId64 ","
1814 "'wr_highest_offset': %" PRId64
1815 "} }",
1816 bs->rd_bytes, bs->wr_bytes,
1817 bs->rd_ops, bs->wr_ops,
1818 bs->wr_highest_sector *
1819 (uint64_t)BDRV_SECTOR_SIZE);
1820 dict = qobject_to_qdict(res);
1822 if (*bs->device_name) {
1823 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1826 if (bs->file) {
1827 QObject *parent = bdrv_info_stats_bs(bs->file);
1828 qdict_put_obj(dict, "parent", parent);
1831 return res;
1834 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1836 QObject *obj;
1837 QList *devices;
1838 BlockDriverState *bs;
1840 devices = qlist_new();
1842 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1843 obj = bdrv_info_stats_bs(bs);
1844 qlist_append_obj(devices, obj);
1847 *ret_data = QOBJECT(devices);
1850 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1852 if (bs->backing_hd && bs->backing_hd->encrypted)
1853 return bs->backing_file;
1854 else if (bs->encrypted)
1855 return bs->filename;
1856 else
1857 return NULL;
1860 void bdrv_get_backing_filename(BlockDriverState *bs,
1861 char *filename, int filename_size)
1863 if (!bs->backing_file) {
1864 pstrcpy(filename, filename_size, "");
1865 } else {
1866 pstrcpy(filename, filename_size, bs->backing_file);
1870 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1871 const uint8_t *buf, int nb_sectors)
1873 BlockDriver *drv = bs->drv;
1874 if (!drv)
1875 return -ENOMEDIUM;
1876 if (!drv->bdrv_write_compressed)
1877 return -ENOTSUP;
1878 if (bdrv_check_request(bs, sector_num, nb_sectors))
1879 return -EIO;
1881 if (bs->dirty_bitmap) {
1882 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1885 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1888 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1890 BlockDriver *drv = bs->drv;
1891 if (!drv)
1892 return -ENOMEDIUM;
1893 if (!drv->bdrv_get_info)
1894 return -ENOTSUP;
1895 memset(bdi, 0, sizeof(*bdi));
1896 return drv->bdrv_get_info(bs, bdi);
1899 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1900 int64_t pos, int size)
1902 BlockDriver *drv = bs->drv;
1903 if (!drv)
1904 return -ENOMEDIUM;
1905 if (drv->bdrv_save_vmstate)
1906 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1907 if (bs->file)
1908 return bdrv_save_vmstate(bs->file, buf, pos, size);
1909 return -ENOTSUP;
1912 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1913 int64_t pos, int size)
1915 BlockDriver *drv = bs->drv;
1916 if (!drv)
1917 return -ENOMEDIUM;
1918 if (drv->bdrv_load_vmstate)
1919 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1920 if (bs->file)
1921 return bdrv_load_vmstate(bs->file, buf, pos, size);
1922 return -ENOTSUP;
1925 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1927 BlockDriver *drv = bs->drv;
1929 if (!drv || !drv->bdrv_debug_event) {
1930 return;
1933 return drv->bdrv_debug_event(bs, event);
1937 /**************************************************************/
1938 /* handling of snapshots */
1940 int bdrv_can_snapshot(BlockDriverState *bs)
1942 BlockDriver *drv = bs->drv;
1943 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1944 return 0;
1947 if (!drv->bdrv_snapshot_create) {
1948 if (bs->file != NULL) {
1949 return bdrv_can_snapshot(bs->file);
1951 return 0;
1954 return 1;
1957 int bdrv_is_snapshot(BlockDriverState *bs)
1959 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1962 BlockDriverState *bdrv_snapshots(void)
1964 BlockDriverState *bs;
1966 if (bs_snapshots) {
1967 return bs_snapshots;
1970 bs = NULL;
1971 while ((bs = bdrv_next(bs))) {
1972 if (bdrv_can_snapshot(bs)) {
1973 bs_snapshots = bs;
1974 return bs;
1977 return NULL;
1980 int bdrv_snapshot_create(BlockDriverState *bs,
1981 QEMUSnapshotInfo *sn_info)
1983 BlockDriver *drv = bs->drv;
1984 if (!drv)
1985 return -ENOMEDIUM;
1986 if (drv->bdrv_snapshot_create)
1987 return drv->bdrv_snapshot_create(bs, sn_info);
1988 if (bs->file)
1989 return bdrv_snapshot_create(bs->file, sn_info);
1990 return -ENOTSUP;
1993 int bdrv_snapshot_goto(BlockDriverState *bs,
1994 const char *snapshot_id)
1996 BlockDriver *drv = bs->drv;
1997 int ret, open_ret;
1999 if (!drv)
2000 return -ENOMEDIUM;
2001 if (drv->bdrv_snapshot_goto)
2002 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2004 if (bs->file) {
2005 drv->bdrv_close(bs);
2006 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2007 open_ret = drv->bdrv_open(bs, bs->open_flags);
2008 if (open_ret < 0) {
2009 bdrv_delete(bs->file);
2010 bs->drv = NULL;
2011 return open_ret;
2013 return ret;
2016 return -ENOTSUP;
2019 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2021 BlockDriver *drv = bs->drv;
2022 if (!drv)
2023 return -ENOMEDIUM;
2024 if (drv->bdrv_snapshot_delete)
2025 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2026 if (bs->file)
2027 return bdrv_snapshot_delete(bs->file, snapshot_id);
2028 return -ENOTSUP;
2031 int bdrv_snapshot_list(BlockDriverState *bs,
2032 QEMUSnapshotInfo **psn_info)
2034 BlockDriver *drv = bs->drv;
2035 if (!drv)
2036 return -ENOMEDIUM;
2037 if (drv->bdrv_snapshot_list)
2038 return drv->bdrv_snapshot_list(bs, psn_info);
2039 if (bs->file)
2040 return bdrv_snapshot_list(bs->file, psn_info);
2041 return -ENOTSUP;
2044 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2045 const char *snapshot_name)
2047 BlockDriver *drv = bs->drv;
2048 if (!drv) {
2049 return -ENOMEDIUM;
2051 if (!bs->read_only) {
2052 return -EINVAL;
2054 if (drv->bdrv_snapshot_load_tmp) {
2055 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2057 return -ENOTSUP;
2060 #define NB_SUFFIXES 4
2062 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2064 static const char suffixes[NB_SUFFIXES] = "KMGT";
2065 int64_t base;
2066 int i;
2068 if (size <= 999) {
2069 snprintf(buf, buf_size, "%" PRId64, size);
2070 } else {
2071 base = 1024;
2072 for(i = 0; i < NB_SUFFIXES; i++) {
2073 if (size < (10 * base)) {
2074 snprintf(buf, buf_size, "%0.1f%c",
2075 (double)size / base,
2076 suffixes[i]);
2077 break;
2078 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2079 snprintf(buf, buf_size, "%" PRId64 "%c",
2080 ((size + (base >> 1)) / base),
2081 suffixes[i]);
2082 break;
2084 base = base * 1024;
2087 return buf;
2090 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2092 char buf1[128], date_buf[128], clock_buf[128];
2093 #ifdef _WIN32
2094 struct tm *ptm;
2095 #else
2096 struct tm tm;
2097 #endif
2098 time_t ti;
2099 int64_t secs;
2101 if (!sn) {
2102 snprintf(buf, buf_size,
2103 "%-10s%-20s%7s%20s%15s",
2104 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2105 } else {
2106 ti = sn->date_sec;
2107 #ifdef _WIN32
2108 ptm = localtime(&ti);
2109 strftime(date_buf, sizeof(date_buf),
2110 "%Y-%m-%d %H:%M:%S", ptm);
2111 #else
2112 localtime_r(&ti, &tm);
2113 strftime(date_buf, sizeof(date_buf),
2114 "%Y-%m-%d %H:%M:%S", &tm);
2115 #endif
2116 secs = sn->vm_clock_nsec / 1000000000;
2117 snprintf(clock_buf, sizeof(clock_buf),
2118 "%02d:%02d:%02d.%03d",
2119 (int)(secs / 3600),
2120 (int)((secs / 60) % 60),
2121 (int)(secs % 60),
2122 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2123 snprintf(buf, buf_size,
2124 "%-10s%-20s%7s%20s%15s",
2125 sn->id_str, sn->name,
2126 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2127 date_buf,
2128 clock_buf);
2130 return buf;
2134 /**************************************************************/
2135 /* async I/Os */
2137 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2138 QEMUIOVector *qiov, int nb_sectors,
2139 BlockDriverCompletionFunc *cb, void *opaque)
2141 BlockDriver *drv = bs->drv;
2142 BlockDriverAIOCB *ret;
2144 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2146 if (!drv)
2147 return NULL;
2148 if (bdrv_check_request(bs, sector_num, nb_sectors))
2149 return NULL;
2151 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2152 cb, opaque);
2154 if (ret) {
2155 /* Update stats even though technically transfer has not happened. */
2156 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2157 bs->rd_ops ++;
2160 return ret;
2163 typedef struct BlockCompleteData {
2164 BlockDriverCompletionFunc *cb;
2165 void *opaque;
2166 BlockDriverState *bs;
2167 int64_t sector_num;
2168 int nb_sectors;
2169 } BlockCompleteData;
2171 static void block_complete_cb(void *opaque, int ret)
2173 BlockCompleteData *b = opaque;
2175 if (b->bs->dirty_bitmap) {
2176 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2178 b->cb(b->opaque, ret);
2179 qemu_free(b);
2182 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2183 int64_t sector_num,
2184 int nb_sectors,
2185 BlockDriverCompletionFunc *cb,
2186 void *opaque)
2188 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2190 blkdata->bs = bs;
2191 blkdata->cb = cb;
2192 blkdata->opaque = opaque;
2193 blkdata->sector_num = sector_num;
2194 blkdata->nb_sectors = nb_sectors;
2196 return blkdata;
2199 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2200 QEMUIOVector *qiov, int nb_sectors,
2201 BlockDriverCompletionFunc *cb, void *opaque)
2203 BlockDriver *drv = bs->drv;
2204 BlockDriverAIOCB *ret;
2205 BlockCompleteData *blk_cb_data;
2207 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2209 if (!drv)
2210 return NULL;
2211 if (bs->read_only)
2212 return NULL;
2213 if (bdrv_check_request(bs, sector_num, nb_sectors))
2214 return NULL;
2216 if (bs->dirty_bitmap) {
2217 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2218 opaque);
2219 cb = &block_complete_cb;
2220 opaque = blk_cb_data;
2223 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2224 cb, opaque);
2226 if (ret) {
2227 /* Update stats even though technically transfer has not happened. */
2228 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2229 bs->wr_ops ++;
2230 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2231 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2235 return ret;
2239 typedef struct MultiwriteCB {
2240 int error;
2241 int num_requests;
2242 int num_callbacks;
2243 struct {
2244 BlockDriverCompletionFunc *cb;
2245 void *opaque;
2246 QEMUIOVector *free_qiov;
2247 void *free_buf;
2248 } callbacks[];
2249 } MultiwriteCB;
2251 static void multiwrite_user_cb(MultiwriteCB *mcb)
2253 int i;
2255 for (i = 0; i < mcb->num_callbacks; i++) {
2256 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2257 if (mcb->callbacks[i].free_qiov) {
2258 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2260 qemu_free(mcb->callbacks[i].free_qiov);
2261 qemu_vfree(mcb->callbacks[i].free_buf);
2265 static void multiwrite_cb(void *opaque, int ret)
2267 MultiwriteCB *mcb = opaque;
2269 trace_multiwrite_cb(mcb, ret);
2271 if (ret < 0 && !mcb->error) {
2272 mcb->error = ret;
2275 mcb->num_requests--;
2276 if (mcb->num_requests == 0) {
2277 multiwrite_user_cb(mcb);
2278 qemu_free(mcb);
2282 static int multiwrite_req_compare(const void *a, const void *b)
2284 const BlockRequest *req1 = a, *req2 = b;
2287 * Note that we can't simply subtract req2->sector from req1->sector
2288 * here as that could overflow the return value.
2290 if (req1->sector > req2->sector) {
2291 return 1;
2292 } else if (req1->sector < req2->sector) {
2293 return -1;
2294 } else {
2295 return 0;
2300 * Takes a bunch of requests and tries to merge them. Returns the number of
2301 * requests that remain after merging.
2303 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2304 int num_reqs, MultiwriteCB *mcb)
2306 int i, outidx;
2308 // Sort requests by start sector
2309 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2311 // Check if adjacent requests touch the same clusters. If so, combine them,
2312 // filling up gaps with zero sectors.
2313 outidx = 0;
2314 for (i = 1; i < num_reqs; i++) {
2315 int merge = 0;
2316 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2318 // This handles the cases that are valid for all block drivers, namely
2319 // exactly sequential writes and overlapping writes.
2320 if (reqs[i].sector <= oldreq_last) {
2321 merge = 1;
2324 // The block driver may decide that it makes sense to combine requests
2325 // even if there is a gap of some sectors between them. In this case,
2326 // the gap is filled with zeros (therefore only applicable for yet
2327 // unused space in format like qcow2).
2328 if (!merge && bs->drv->bdrv_merge_requests) {
2329 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2332 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2333 merge = 0;
2336 if (merge) {
2337 size_t size;
2338 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2339 qemu_iovec_init(qiov,
2340 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2342 // Add the first request to the merged one. If the requests are
2343 // overlapping, drop the last sectors of the first request.
2344 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2345 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2347 // We might need to add some zeros between the two requests
2348 if (reqs[i].sector > oldreq_last) {
2349 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2350 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2351 memset(buf, 0, zero_bytes);
2352 qemu_iovec_add(qiov, buf, zero_bytes);
2353 mcb->callbacks[i].free_buf = buf;
2356 // Add the second request
2357 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2359 reqs[outidx].nb_sectors = qiov->size >> 9;
2360 reqs[outidx].qiov = qiov;
2362 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2363 } else {
2364 outidx++;
2365 reqs[outidx].sector = reqs[i].sector;
2366 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2367 reqs[outidx].qiov = reqs[i].qiov;
2371 return outidx + 1;
2375 * Submit multiple AIO write requests at once.
2377 * On success, the function returns 0 and all requests in the reqs array have
2378 * been submitted. In error case this function returns -1, and any of the
2379 * requests may or may not be submitted yet. In particular, this means that the
2380 * callback will be called for some of the requests, for others it won't. The
2381 * caller must check the error field of the BlockRequest to wait for the right
2382 * callbacks (if error != 0, no callback will be called).
2384 * The implementation may modify the contents of the reqs array, e.g. to merge
2385 * requests. However, the fields opaque and error are left unmodified as they
2386 * are used to signal failure for a single request to the caller.
2388 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2390 BlockDriverAIOCB *acb;
2391 MultiwriteCB *mcb;
2392 int i;
2394 /* don't submit writes if we don't have a medium */
2395 if (bs->drv == NULL) {
2396 for (i = 0; i < num_reqs; i++) {
2397 reqs[i].error = -ENOMEDIUM;
2399 return -1;
2402 if (num_reqs == 0) {
2403 return 0;
2406 // Create MultiwriteCB structure
2407 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2408 mcb->num_requests = 0;
2409 mcb->num_callbacks = num_reqs;
2411 for (i = 0; i < num_reqs; i++) {
2412 mcb->callbacks[i].cb = reqs[i].cb;
2413 mcb->callbacks[i].opaque = reqs[i].opaque;
2416 // Check for mergable requests
2417 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2419 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2422 * Run the aio requests. As soon as one request can't be submitted
2423 * successfully, fail all requests that are not yet submitted (we must
2424 * return failure for all requests anyway)
2426 * num_requests cannot be set to the right value immediately: If
2427 * bdrv_aio_writev fails for some request, num_requests would be too high
2428 * and therefore multiwrite_cb() would never recognize the multiwrite
2429 * request as completed. We also cannot use the loop variable i to set it
2430 * when the first request fails because the callback may already have been
2431 * called for previously submitted requests. Thus, num_requests must be
2432 * incremented for each request that is submitted.
2434 * The problem that callbacks may be called early also means that we need
2435 * to take care that num_requests doesn't become 0 before all requests are
2436 * submitted - multiwrite_cb() would consider the multiwrite request
2437 * completed. A dummy request that is "completed" by a manual call to
2438 * multiwrite_cb() takes care of this.
2440 mcb->num_requests = 1;
2442 // Run the aio requests
2443 for (i = 0; i < num_reqs; i++) {
2444 mcb->num_requests++;
2445 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2446 reqs[i].nb_sectors, multiwrite_cb, mcb);
2448 if (acb == NULL) {
2449 // We can only fail the whole thing if no request has been
2450 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2451 // complete and report the error in the callback.
2452 if (i == 0) {
2453 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2454 goto fail;
2455 } else {
2456 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2457 multiwrite_cb(mcb, -EIO);
2458 break;
2463 /* Complete the dummy request */
2464 multiwrite_cb(mcb, 0);
2466 return 0;
2468 fail:
2469 for (i = 0; i < mcb->num_callbacks; i++) {
2470 reqs[i].error = -EIO;
2472 qemu_free(mcb);
2473 return -1;
2476 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2477 BlockDriverCompletionFunc *cb, void *opaque)
2479 BlockDriver *drv = bs->drv;
2481 trace_bdrv_aio_flush(bs, opaque);
2483 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2484 return bdrv_aio_noop_em(bs, cb, opaque);
2487 if (!drv)
2488 return NULL;
2489 return drv->bdrv_aio_flush(bs, cb, opaque);
2492 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2494 acb->pool->cancel(acb);
2498 /**************************************************************/
2499 /* async block device emulation */
2501 typedef struct BlockDriverAIOCBSync {
2502 BlockDriverAIOCB common;
2503 QEMUBH *bh;
2504 int ret;
2505 /* vector translation state */
2506 QEMUIOVector *qiov;
2507 uint8_t *bounce;
2508 int is_write;
2509 } BlockDriverAIOCBSync;
2511 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2513 BlockDriverAIOCBSync *acb =
2514 container_of(blockacb, BlockDriverAIOCBSync, common);
2515 qemu_bh_delete(acb->bh);
2516 acb->bh = NULL;
2517 qemu_aio_release(acb);
2520 static AIOPool bdrv_em_aio_pool = {
2521 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2522 .cancel = bdrv_aio_cancel_em,
2525 static void bdrv_aio_bh_cb(void *opaque)
2527 BlockDriverAIOCBSync *acb = opaque;
2529 if (!acb->is_write)
2530 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2531 qemu_vfree(acb->bounce);
2532 acb->common.cb(acb->common.opaque, acb->ret);
2533 qemu_bh_delete(acb->bh);
2534 acb->bh = NULL;
2535 qemu_aio_release(acb);
2538 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2539 int64_t sector_num,
2540 QEMUIOVector *qiov,
2541 int nb_sectors,
2542 BlockDriverCompletionFunc *cb,
2543 void *opaque,
2544 int is_write)
2547 BlockDriverAIOCBSync *acb;
2549 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2550 acb->is_write = is_write;
2551 acb->qiov = qiov;
2552 acb->bounce = qemu_blockalign(bs, qiov->size);
2554 if (!acb->bh)
2555 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2557 if (is_write) {
2558 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2559 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2560 } else {
2561 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2564 qemu_bh_schedule(acb->bh);
2566 return &acb->common;
2569 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2570 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2571 BlockDriverCompletionFunc *cb, void *opaque)
2573 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2576 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2577 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2578 BlockDriverCompletionFunc *cb, void *opaque)
2580 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2583 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2584 BlockDriverCompletionFunc *cb, void *opaque)
2586 BlockDriverAIOCBSync *acb;
2588 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2589 acb->is_write = 1; /* don't bounce in the completion hadler */
2590 acb->qiov = NULL;
2591 acb->bounce = NULL;
2592 acb->ret = 0;
2594 if (!acb->bh)
2595 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2597 bdrv_flush(bs);
2598 qemu_bh_schedule(acb->bh);
2599 return &acb->common;
2602 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2603 BlockDriverCompletionFunc *cb, void *opaque)
2605 BlockDriverAIOCBSync *acb;
2607 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2608 acb->is_write = 1; /* don't bounce in the completion handler */
2609 acb->qiov = NULL;
2610 acb->bounce = NULL;
2611 acb->ret = 0;
2613 if (!acb->bh) {
2614 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2617 qemu_bh_schedule(acb->bh);
2618 return &acb->common;
2621 /**************************************************************/
2622 /* sync block device emulation */
2624 static void bdrv_rw_em_cb(void *opaque, int ret)
2626 *(int *)opaque = ret;
2629 #define NOT_DONE 0x7fffffff
2631 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2632 uint8_t *buf, int nb_sectors)
2634 int async_ret;
2635 BlockDriverAIOCB *acb;
2636 struct iovec iov;
2637 QEMUIOVector qiov;
2639 async_context_push();
2641 async_ret = NOT_DONE;
2642 iov.iov_base = (void *)buf;
2643 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2644 qemu_iovec_init_external(&qiov, &iov, 1);
2645 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2646 bdrv_rw_em_cb, &async_ret);
2647 if (acb == NULL) {
2648 async_ret = -1;
2649 goto fail;
2652 while (async_ret == NOT_DONE) {
2653 qemu_aio_wait();
2657 fail:
2658 async_context_pop();
2659 return async_ret;
2662 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2663 const uint8_t *buf, int nb_sectors)
2665 int async_ret;
2666 BlockDriverAIOCB *acb;
2667 struct iovec iov;
2668 QEMUIOVector qiov;
2670 async_context_push();
2672 async_ret = NOT_DONE;
2673 iov.iov_base = (void *)buf;
2674 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2675 qemu_iovec_init_external(&qiov, &iov, 1);
2676 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2677 bdrv_rw_em_cb, &async_ret);
2678 if (acb == NULL) {
2679 async_ret = -1;
2680 goto fail;
2682 while (async_ret == NOT_DONE) {
2683 qemu_aio_wait();
2686 fail:
2687 async_context_pop();
2688 return async_ret;
2691 void bdrv_init(void)
2693 module_call_init(MODULE_INIT_BLOCK);
2696 void bdrv_init_with_whitelist(void)
2698 use_bdrv_whitelist = 1;
2699 bdrv_init();
2702 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2703 BlockDriverCompletionFunc *cb, void *opaque)
2705 BlockDriverAIOCB *acb;
2707 if (pool->free_aiocb) {
2708 acb = pool->free_aiocb;
2709 pool->free_aiocb = acb->next;
2710 } else {
2711 acb = qemu_mallocz(pool->aiocb_size);
2712 acb->pool = pool;
2714 acb->bs = bs;
2715 acb->cb = cb;
2716 acb->opaque = opaque;
2717 return acb;
2720 void qemu_aio_release(void *p)
2722 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2723 AIOPool *pool = acb->pool;
2724 acb->next = pool->free_aiocb;
2725 pool->free_aiocb = acb;
2728 /**************************************************************/
2729 /* removable device support */
2732 * Return TRUE if the media is present
2734 int bdrv_is_inserted(BlockDriverState *bs)
2736 BlockDriver *drv = bs->drv;
2737 int ret;
2738 if (!drv)
2739 return 0;
2740 if (!drv->bdrv_is_inserted)
2741 return !bs->tray_open;
2742 ret = drv->bdrv_is_inserted(bs);
2743 return ret;
2747 * Return TRUE if the media changed since the last call to this
2748 * function. It is currently only used for floppy disks
2750 int bdrv_media_changed(BlockDriverState *bs)
2752 BlockDriver *drv = bs->drv;
2753 int ret;
2755 if (!drv || !drv->bdrv_media_changed)
2756 ret = -ENOTSUP;
2757 else
2758 ret = drv->bdrv_media_changed(bs);
2759 if (ret == -ENOTSUP)
2760 ret = bs->media_changed;
2761 bs->media_changed = 0;
2762 return ret;
2766 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2768 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2770 BlockDriver *drv = bs->drv;
2771 int ret;
2773 if (bs->locked) {
2774 return -EBUSY;
2777 if (!drv || !drv->bdrv_eject) {
2778 ret = -ENOTSUP;
2779 } else {
2780 ret = drv->bdrv_eject(bs, eject_flag);
2782 if (ret == -ENOTSUP) {
2783 ret = 0;
2785 if (ret >= 0) {
2786 bs->tray_open = eject_flag;
2789 return ret;
2792 int bdrv_is_locked(BlockDriverState *bs)
2794 return bs->locked;
2798 * Lock or unlock the media (if it is locked, the user won't be able
2799 * to eject it manually).
2801 void bdrv_set_locked(BlockDriverState *bs, int locked)
2803 BlockDriver *drv = bs->drv;
2805 trace_bdrv_set_locked(bs, locked);
2807 bs->locked = locked;
2808 if (drv && drv->bdrv_set_locked) {
2809 drv->bdrv_set_locked(bs, locked);
2813 /* needed for generic scsi interface */
2815 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2817 BlockDriver *drv = bs->drv;
2819 if (drv && drv->bdrv_ioctl)
2820 return drv->bdrv_ioctl(bs, req, buf);
2821 return -ENOTSUP;
2824 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2825 unsigned long int req, void *buf,
2826 BlockDriverCompletionFunc *cb, void *opaque)
2828 BlockDriver *drv = bs->drv;
2830 if (drv && drv->bdrv_aio_ioctl)
2831 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2832 return NULL;
2837 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2839 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2842 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2844 int64_t bitmap_size;
2846 bs->dirty_count = 0;
2847 if (enable) {
2848 if (!bs->dirty_bitmap) {
2849 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2850 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2851 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2853 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2855 } else {
2856 if (bs->dirty_bitmap) {
2857 qemu_free(bs->dirty_bitmap);
2858 bs->dirty_bitmap = NULL;
2863 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2865 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2867 if (bs->dirty_bitmap &&
2868 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2869 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2870 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2871 } else {
2872 return 0;
2876 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2877 int nr_sectors)
2879 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2882 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2884 return bs->dirty_count;
2887 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
2889 assert(bs->in_use != in_use);
2890 bs->in_use = in_use;
2893 int bdrv_in_use(BlockDriverState *bs)
2895 return bs->in_use;
2898 int bdrv_img_create(const char *filename, const char *fmt,
2899 const char *base_filename, const char *base_fmt,
2900 char *options, uint64_t img_size, int flags)
2902 QEMUOptionParameter *param = NULL, *create_options = NULL;
2903 QEMUOptionParameter *backing_fmt, *backing_file, *size;
2904 BlockDriverState *bs = NULL;
2905 BlockDriver *drv, *proto_drv;
2906 BlockDriver *backing_drv = NULL;
2907 int ret = 0;
2909 /* Find driver and parse its options */
2910 drv = bdrv_find_format(fmt);
2911 if (!drv) {
2912 error_report("Unknown file format '%s'", fmt);
2913 ret = -EINVAL;
2914 goto out;
2917 proto_drv = bdrv_find_protocol(filename);
2918 if (!proto_drv) {
2919 error_report("Unknown protocol '%s'", filename);
2920 ret = -EINVAL;
2921 goto out;
2924 create_options = append_option_parameters(create_options,
2925 drv->create_options);
2926 create_options = append_option_parameters(create_options,
2927 proto_drv->create_options);
2929 /* Create parameter list with default values */
2930 param = parse_option_parameters("", create_options, param);
2932 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
2934 /* Parse -o options */
2935 if (options) {
2936 param = parse_option_parameters(options, create_options, param);
2937 if (param == NULL) {
2938 error_report("Invalid options for file format '%s'.", fmt);
2939 ret = -EINVAL;
2940 goto out;
2944 if (base_filename) {
2945 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
2946 base_filename)) {
2947 error_report("Backing file not supported for file format '%s'",
2948 fmt);
2949 ret = -EINVAL;
2950 goto out;
2954 if (base_fmt) {
2955 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
2956 error_report("Backing file format not supported for file "
2957 "format '%s'", fmt);
2958 ret = -EINVAL;
2959 goto out;
2963 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
2964 if (backing_file && backing_file->value.s) {
2965 if (!strcmp(filename, backing_file->value.s)) {
2966 error_report("Error: Trying to create an image with the "
2967 "same filename as the backing file");
2968 ret = -EINVAL;
2969 goto out;
2973 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
2974 if (backing_fmt && backing_fmt->value.s) {
2975 backing_drv = bdrv_find_format(backing_fmt->value.s);
2976 if (!backing_drv) {
2977 error_report("Unknown backing file format '%s'",
2978 backing_fmt->value.s);
2979 ret = -EINVAL;
2980 goto out;
2984 // The size for the image must always be specified, with one exception:
2985 // If we are using a backing file, we can obtain the size from there
2986 size = get_option_parameter(param, BLOCK_OPT_SIZE);
2987 if (size && size->value.n == -1) {
2988 if (backing_file && backing_file->value.s) {
2989 uint64_t size;
2990 char buf[32];
2992 bs = bdrv_new("");
2994 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
2995 if (ret < 0) {
2996 error_report("Could not open '%s'", backing_file->value.s);
2997 goto out;
2999 bdrv_get_geometry(bs, &size);
3000 size *= 512;
3002 snprintf(buf, sizeof(buf), "%" PRId64, size);
3003 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3004 } else {
3005 error_report("Image creation needs a size parameter");
3006 ret = -EINVAL;
3007 goto out;
3011 printf("Formatting '%s', fmt=%s ", filename, fmt);
3012 print_option_parameters(param);
3013 puts("");
3015 ret = bdrv_create(drv, filename, param);
3017 if (ret < 0) {
3018 if (ret == -ENOTSUP) {
3019 error_report("Formatting or formatting option not supported for "
3020 "file format '%s'", fmt);
3021 } else if (ret == -EFBIG) {
3022 error_report("The image size is too large for file format '%s'",
3023 fmt);
3024 } else {
3025 error_report("%s: error while creating %s: %s", filename, fmt,
3026 strerror(-ret));
3030 out:
3031 free_option_parameters(create_options);
3032 free_option_parameters(param);
3034 if (bs) {
3035 bdrv_delete(bs);
3038 return ret;