s390: typo fix
[qemu/agraf.git] / block.c
blob3f3496ec90ad02e90365074ae9b06085de1c499e
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 "monitor.h"
27 #include "block_int.h"
28 #include "module.h"
29 #include "qemu-objects.h"
31 #ifdef CONFIG_BSD
32 #include <sys/types.h>
33 #include <sys/stat.h>
34 #include <sys/ioctl.h>
35 #include <sys/queue.h>
36 #ifndef __DragonFly__
37 #include <sys/disk.h>
38 #endif
39 #endif
41 #ifdef _WIN32
42 #include <windows.h>
43 #endif
45 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
46 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
47 BlockDriverCompletionFunc *cb, void *opaque);
48 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
49 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
50 BlockDriverCompletionFunc *cb, void *opaque);
51 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
52 BlockDriverCompletionFunc *cb, void *opaque);
53 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
54 uint8_t *buf, int nb_sectors);
55 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
56 const uint8_t *buf, int nb_sectors);
58 BlockDriverState *bdrv_first;
60 static BlockDriver *first_drv;
62 /* If non-zero, use only whitelisted block drivers */
63 static int use_bdrv_whitelist;
65 int path_is_absolute(const char *path)
67 const char *p;
68 #ifdef _WIN32
69 /* specific case for names like: "\\.\d:" */
70 if (*path == '/' || *path == '\\')
71 return 1;
72 #endif
73 p = strchr(path, ':');
74 if (p)
75 p++;
76 else
77 p = path;
78 #ifdef _WIN32
79 return (*p == '/' || *p == '\\');
80 #else
81 return (*p == '/');
82 #endif
85 /* if filename is absolute, just copy it to dest. Otherwise, build a
86 path to it by considering it is relative to base_path. URL are
87 supported. */
88 void path_combine(char *dest, int dest_size,
89 const char *base_path,
90 const char *filename)
92 const char *p, *p1;
93 int len;
95 if (dest_size <= 0)
96 return;
97 if (path_is_absolute(filename)) {
98 pstrcpy(dest, dest_size, filename);
99 } else {
100 p = strchr(base_path, ':');
101 if (p)
102 p++;
103 else
104 p = base_path;
105 p1 = strrchr(base_path, '/');
106 #ifdef _WIN32
108 const char *p2;
109 p2 = strrchr(base_path, '\\');
110 if (!p1 || p2 > p1)
111 p1 = p2;
113 #endif
114 if (p1)
115 p1++;
116 else
117 p1 = base_path;
118 if (p1 > p)
119 p = p1;
120 len = p - base_path;
121 if (len > dest_size - 1)
122 len = dest_size - 1;
123 memcpy(dest, base_path, len);
124 dest[len] = '\0';
125 pstrcat(dest, dest_size, filename);
129 void bdrv_register(BlockDriver *bdrv)
131 if (!bdrv->bdrv_aio_readv) {
132 /* add AIO emulation layer */
133 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
134 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
135 } else if (!bdrv->bdrv_read) {
136 /* add synchronous IO emulation layer */
137 bdrv->bdrv_read = bdrv_read_em;
138 bdrv->bdrv_write = bdrv_write_em;
141 if (!bdrv->bdrv_aio_flush)
142 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
144 bdrv->next = first_drv;
145 first_drv = bdrv;
148 /* create a new block device (by default it is empty) */
149 BlockDriverState *bdrv_new(const char *device_name)
151 BlockDriverState **pbs, *bs;
153 bs = qemu_mallocz(sizeof(BlockDriverState));
154 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
155 if (device_name[0] != '\0') {
156 /* insert at the end */
157 pbs = &bdrv_first;
158 while (*pbs != NULL)
159 pbs = &(*pbs)->next;
160 *pbs = bs;
162 return bs;
165 BlockDriver *bdrv_find_format(const char *format_name)
167 BlockDriver *drv1;
168 for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
169 if (!strcmp(drv1->format_name, format_name))
170 return drv1;
172 return NULL;
175 static int bdrv_is_whitelisted(BlockDriver *drv)
177 static const char *whitelist[] = {
178 CONFIG_BDRV_WHITELIST
180 const char **p;
182 if (!whitelist[0])
183 return 1; /* no whitelist, anything goes */
185 for (p = whitelist; *p; p++) {
186 if (!strcmp(drv->format_name, *p)) {
187 return 1;
190 return 0;
193 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
195 BlockDriver *drv = bdrv_find_format(format_name);
196 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
199 int bdrv_create(BlockDriver *drv, const char* filename,
200 QEMUOptionParameter *options)
202 if (!drv->bdrv_create)
203 return -ENOTSUP;
205 return drv->bdrv_create(filename, options);
208 #ifdef _WIN32
209 void get_tmp_filename(char *filename, int size)
211 char temp_dir[MAX_PATH];
213 GetTempPath(MAX_PATH, temp_dir);
214 GetTempFileName(temp_dir, "qem", 0, filename);
216 #else
217 void get_tmp_filename(char *filename, int size)
219 int fd;
220 const char *tmpdir;
221 /* XXX: race condition possible */
222 tmpdir = getenv("TMPDIR");
223 if (!tmpdir)
224 tmpdir = "/tmp";
225 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
226 fd = mkstemp(filename);
227 close(fd);
229 #endif
231 #ifdef _WIN32
232 static int is_windows_drive_prefix(const char *filename)
234 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
235 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
236 filename[1] == ':');
239 int is_windows_drive(const char *filename)
241 if (is_windows_drive_prefix(filename) &&
242 filename[2] == '\0')
243 return 1;
244 if (strstart(filename, "\\\\.\\", NULL) ||
245 strstart(filename, "//./", NULL))
246 return 1;
247 return 0;
249 #endif
251 static BlockDriver *find_protocol(const char *filename)
253 BlockDriver *drv1;
254 char protocol[128];
255 int len;
256 const char *p;
258 #ifdef _WIN32
259 if (is_windows_drive(filename) ||
260 is_windows_drive_prefix(filename))
261 return bdrv_find_format("raw");
262 #endif
263 p = strchr(filename, ':');
264 if (!p)
265 return bdrv_find_format("raw");
266 len = p - filename;
267 if (len > sizeof(protocol) - 1)
268 len = sizeof(protocol) - 1;
269 memcpy(protocol, filename, len);
270 protocol[len] = '\0';
271 for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
272 if (drv1->protocol_name &&
273 !strcmp(drv1->protocol_name, protocol))
274 return drv1;
276 return NULL;
280 * Detect host devices. By convention, /dev/cdrom[N] is always
281 * recognized as a host CDROM.
283 static BlockDriver *find_hdev_driver(const char *filename)
285 int score_max = 0, score;
286 BlockDriver *drv = NULL, *d;
288 for (d = first_drv; d; d = d->next) {
289 if (d->bdrv_probe_device) {
290 score = d->bdrv_probe_device(filename);
291 if (score > score_max) {
292 score_max = score;
293 drv = d;
298 return drv;
301 static BlockDriver *find_image_format(const char *filename)
303 int ret, score, score_max;
304 BlockDriver *drv1, *drv;
305 uint8_t buf[2048];
306 BlockDriverState *bs;
308 drv = find_protocol(filename);
309 /* no need to test disk image formats for vvfat */
310 if (drv && strcmp(drv->format_name, "vvfat") == 0)
311 return drv;
313 ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
314 if (ret < 0)
315 return NULL;
316 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
317 bdrv_delete(bs);
318 if (ret < 0) {
319 return NULL;
322 score_max = 0;
323 for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
324 if (drv1->bdrv_probe) {
325 score = drv1->bdrv_probe(buf, ret, filename);
326 if (score > score_max) {
327 score_max = score;
328 drv = drv1;
332 return drv;
335 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
337 BlockDriverState *bs;
338 int ret;
340 bs = bdrv_new("");
341 ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
342 if (ret < 0) {
343 bdrv_delete(bs);
344 return ret;
346 bs->growable = 1;
347 *pbs = bs;
348 return 0;
351 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
353 return bdrv_open2(bs, filename, flags, NULL);
356 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
357 BlockDriver *drv)
359 int ret, open_flags, try_rw;
360 char tmp_filename[PATH_MAX];
361 char backing_filename[PATH_MAX];
363 bs->is_temporary = 0;
364 bs->encrypted = 0;
365 bs->valid_key = 0;
366 /* buffer_alignment defaulted to 512, drivers can change this value */
367 bs->buffer_alignment = 512;
369 if (flags & BDRV_O_SNAPSHOT) {
370 BlockDriverState *bs1;
371 int64_t total_size;
372 int is_protocol = 0;
373 BlockDriver *bdrv_qcow2;
374 QEMUOptionParameter *options;
376 /* if snapshot, we create a temporary backing file and open it
377 instead of opening 'filename' directly */
379 /* if there is a backing file, use it */
380 bs1 = bdrv_new("");
381 ret = bdrv_open2(bs1, filename, 0, drv);
382 if (ret < 0) {
383 bdrv_delete(bs1);
384 return ret;
386 total_size = bdrv_getlength(bs1) >> BDRV_SECTOR_BITS;
388 if (bs1->drv && bs1->drv->protocol_name)
389 is_protocol = 1;
391 bdrv_delete(bs1);
393 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
395 /* Real path is meaningless for protocols */
396 if (is_protocol)
397 snprintf(backing_filename, sizeof(backing_filename),
398 "%s", filename);
399 else
400 realpath(filename, backing_filename);
402 bdrv_qcow2 = bdrv_find_format("qcow2");
403 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
405 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
406 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
407 if (drv) {
408 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
409 drv->format_name);
412 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
413 if (ret < 0) {
414 return ret;
417 filename = tmp_filename;
418 drv = bdrv_qcow2;
419 bs->is_temporary = 1;
422 pstrcpy(bs->filename, sizeof(bs->filename), filename);
423 if (flags & BDRV_O_FILE) {
424 drv = find_protocol(filename);
425 } else if (!drv) {
426 drv = find_hdev_driver(filename);
427 if (!drv) {
428 drv = find_image_format(filename);
431 if (!drv) {
432 ret = -ENOENT;
433 goto unlink_and_fail;
435 bs->drv = drv;
436 bs->opaque = qemu_mallocz(drv->instance_size);
439 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
440 * write cache to the guest. We do need the fdatasync to flush
441 * out transactions for block allocations, and we maybe have a
442 * volatile write cache in our backing device to deal with.
444 if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
445 bs->enable_write_cache = 1;
447 /* Note: for compatibility, we open disk image files as RDWR, and
448 RDONLY as fallback */
449 try_rw = !bs->read_only || bs->is_temporary;
450 if (!(flags & BDRV_O_FILE))
451 open_flags = (try_rw ? BDRV_O_RDWR : 0) |
452 (flags & (BDRV_O_CACHE_MASK|BDRV_O_NATIVE_AIO));
453 else
454 open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
455 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv))
456 ret = -ENOTSUP;
457 else
458 ret = drv->bdrv_open(bs, filename, open_flags);
459 if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
460 ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
461 bs->read_only = 1;
463 if (ret < 0) {
464 qemu_free(bs->opaque);
465 bs->opaque = NULL;
466 bs->drv = NULL;
467 unlink_and_fail:
468 if (bs->is_temporary)
469 unlink(filename);
470 return ret;
472 if (drv->bdrv_getlength) {
473 bs->total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
475 #ifndef _WIN32
476 if (bs->is_temporary) {
477 unlink(filename);
479 #endif
480 if (bs->backing_file[0] != '\0') {
481 /* if there is a backing file, use it */
482 BlockDriver *back_drv = NULL;
483 bs->backing_hd = bdrv_new("");
484 /* pass on read_only property to the backing_hd */
485 bs->backing_hd->read_only = bs->read_only;
486 path_combine(backing_filename, sizeof(backing_filename),
487 filename, bs->backing_file);
488 if (bs->backing_format[0] != '\0')
489 back_drv = bdrv_find_format(bs->backing_format);
490 ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
491 back_drv);
492 if (ret < 0) {
493 bdrv_close(bs);
494 return ret;
498 if (!bdrv_key_required(bs)) {
499 /* call the change callback */
500 bs->media_changed = 1;
501 if (bs->change_cb)
502 bs->change_cb(bs->change_opaque);
504 return 0;
507 void bdrv_close(BlockDriverState *bs)
509 if (bs->drv) {
510 if (bs->backing_hd)
511 bdrv_delete(bs->backing_hd);
512 bs->drv->bdrv_close(bs);
513 qemu_free(bs->opaque);
514 #ifdef _WIN32
515 if (bs->is_temporary) {
516 unlink(bs->filename);
518 #endif
519 bs->opaque = NULL;
520 bs->drv = NULL;
522 /* call the change callback */
523 bs->media_changed = 1;
524 if (bs->change_cb)
525 bs->change_cb(bs->change_opaque);
529 void bdrv_delete(BlockDriverState *bs)
531 BlockDriverState **pbs;
533 pbs = &bdrv_first;
534 while (*pbs != bs && *pbs != NULL)
535 pbs = &(*pbs)->next;
536 if (*pbs == bs)
537 *pbs = bs->next;
539 bdrv_close(bs);
540 qemu_free(bs);
544 * Run consistency checks on an image
546 * Returns the number of errors or -errno when an internal error occurs
548 int bdrv_check(BlockDriverState *bs)
550 if (bs->drv->bdrv_check == NULL) {
551 return -ENOTSUP;
554 return bs->drv->bdrv_check(bs);
557 /* commit COW file into the raw image */
558 int bdrv_commit(BlockDriverState *bs)
560 BlockDriver *drv = bs->drv;
561 int64_t i, total_sectors;
562 int n, j;
563 unsigned char sector[512];
565 if (!drv)
566 return -ENOMEDIUM;
568 if (bs->read_only) {
569 return -EACCES;
572 if (!bs->backing_hd) {
573 return -ENOTSUP;
576 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
577 for (i = 0; i < total_sectors;) {
578 if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
579 for(j = 0; j < n; j++) {
580 if (bdrv_read(bs, i, sector, 1) != 0) {
581 return -EIO;
584 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
585 return -EIO;
587 i++;
589 } else {
590 i += n;
594 if (drv->bdrv_make_empty)
595 return drv->bdrv_make_empty(bs);
597 return 0;
600 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
601 size_t size)
603 int64_t len;
605 if (!bdrv_is_inserted(bs))
606 return -ENOMEDIUM;
608 if (bs->growable)
609 return 0;
611 len = bdrv_getlength(bs);
613 if (offset < 0)
614 return -EIO;
616 if ((offset > len) || (len - offset < size))
617 return -EIO;
619 return 0;
622 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
623 int nb_sectors)
625 return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
628 /* return < 0 if error. See bdrv_write() for the return codes */
629 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
630 uint8_t *buf, int nb_sectors)
632 BlockDriver *drv = bs->drv;
634 if (!drv)
635 return -ENOMEDIUM;
636 if (bdrv_check_request(bs, sector_num, nb_sectors))
637 return -EIO;
639 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
642 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
643 int nb_sectors, int dirty)
645 int64_t start, end;
646 unsigned long val, idx, bit;
648 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
649 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
651 for (; start <= end; start++) {
652 idx = start / (sizeof(unsigned long) * 8);
653 bit = start % (sizeof(unsigned long) * 8);
654 val = bs->dirty_bitmap[idx];
655 if (dirty) {
656 val |= 1 << bit;
657 } else {
658 val &= ~(1 << bit);
660 bs->dirty_bitmap[idx] = val;
664 /* Return < 0 if error. Important errors are:
665 -EIO generic I/O error (may happen for all errors)
666 -ENOMEDIUM No media inserted.
667 -EINVAL Invalid sector number or nb_sectors
668 -EACCES Trying to write a read-only device
670 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
671 const uint8_t *buf, int nb_sectors)
673 BlockDriver *drv = bs->drv;
674 if (!bs->drv)
675 return -ENOMEDIUM;
676 if (bs->read_only)
677 return -EACCES;
678 if (bdrv_check_request(bs, sector_num, nb_sectors))
679 return -EIO;
681 if (bs->dirty_bitmap) {
682 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
685 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
688 int bdrv_pread(BlockDriverState *bs, int64_t offset,
689 void *buf, int count1)
691 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
692 int len, nb_sectors, count;
693 int64_t sector_num;
695 count = count1;
696 /* first read to align to sector start */
697 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
698 if (len > count)
699 len = count;
700 sector_num = offset >> BDRV_SECTOR_BITS;
701 if (len > 0) {
702 if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
703 return -EIO;
704 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
705 count -= len;
706 if (count == 0)
707 return count1;
708 sector_num++;
709 buf += len;
712 /* read the sectors "in place" */
713 nb_sectors = count >> BDRV_SECTOR_BITS;
714 if (nb_sectors > 0) {
715 if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
716 return -EIO;
717 sector_num += nb_sectors;
718 len = nb_sectors << BDRV_SECTOR_BITS;
719 buf += len;
720 count -= len;
723 /* add data from the last sector */
724 if (count > 0) {
725 if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
726 return -EIO;
727 memcpy(buf, tmp_buf, count);
729 return count1;
732 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
733 const void *buf, int count1)
735 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
736 int len, nb_sectors, count;
737 int64_t sector_num;
739 count = count1;
740 /* first write to align to sector start */
741 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
742 if (len > count)
743 len = count;
744 sector_num = offset >> BDRV_SECTOR_BITS;
745 if (len > 0) {
746 if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
747 return -EIO;
748 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
749 if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
750 return -EIO;
751 count -= len;
752 if (count == 0)
753 return count1;
754 sector_num++;
755 buf += len;
758 /* write the sectors "in place" */
759 nb_sectors = count >> BDRV_SECTOR_BITS;
760 if (nb_sectors > 0) {
761 if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
762 return -EIO;
763 sector_num += nb_sectors;
764 len = nb_sectors << BDRV_SECTOR_BITS;
765 buf += len;
766 count -= len;
769 /* add data from the last sector */
770 if (count > 0) {
771 if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
772 return -EIO;
773 memcpy(tmp_buf, buf, count);
774 if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
775 return -EIO;
777 return count1;
781 * Truncate file to 'offset' bytes (needed only for file protocols)
783 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
785 BlockDriver *drv = bs->drv;
786 if (!drv)
787 return -ENOMEDIUM;
788 if (!drv->bdrv_truncate)
789 return -ENOTSUP;
790 if (bs->read_only)
791 return -EACCES;
792 return drv->bdrv_truncate(bs, offset);
796 * Length of a file in bytes. Return < 0 if error or unknown.
798 int64_t bdrv_getlength(BlockDriverState *bs)
800 BlockDriver *drv = bs->drv;
801 if (!drv)
802 return -ENOMEDIUM;
803 if (!drv->bdrv_getlength) {
804 /* legacy mode */
805 return bs->total_sectors * BDRV_SECTOR_SIZE;
807 return drv->bdrv_getlength(bs);
810 /* return 0 as number of sectors if no device present or error */
811 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
813 int64_t length;
814 length = bdrv_getlength(bs);
815 if (length < 0)
816 length = 0;
817 else
818 length = length >> BDRV_SECTOR_BITS;
819 *nb_sectors_ptr = length;
822 struct partition {
823 uint8_t boot_ind; /* 0x80 - active */
824 uint8_t head; /* starting head */
825 uint8_t sector; /* starting sector */
826 uint8_t cyl; /* starting cylinder */
827 uint8_t sys_ind; /* What partition type */
828 uint8_t end_head; /* end head */
829 uint8_t end_sector; /* end sector */
830 uint8_t end_cyl; /* end cylinder */
831 uint32_t start_sect; /* starting sector counting from 0 */
832 uint32_t nr_sects; /* nr of sectors in partition */
833 } __attribute__((packed));
835 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
836 static int guess_disk_lchs(BlockDriverState *bs,
837 int *pcylinders, int *pheads, int *psectors)
839 uint8_t buf[512];
840 int ret, i, heads, sectors, cylinders;
841 struct partition *p;
842 uint32_t nr_sects;
843 uint64_t nb_sectors;
845 bdrv_get_geometry(bs, &nb_sectors);
847 ret = bdrv_read(bs, 0, buf, 1);
848 if (ret < 0)
849 return -1;
850 /* test msdos magic */
851 if (buf[510] != 0x55 || buf[511] != 0xaa)
852 return -1;
853 for(i = 0; i < 4; i++) {
854 p = ((struct partition *)(buf + 0x1be)) + i;
855 nr_sects = le32_to_cpu(p->nr_sects);
856 if (nr_sects && p->end_head) {
857 /* We make the assumption that the partition terminates on
858 a cylinder boundary */
859 heads = p->end_head + 1;
860 sectors = p->end_sector & 63;
861 if (sectors == 0)
862 continue;
863 cylinders = nb_sectors / (heads * sectors);
864 if (cylinders < 1 || cylinders > 16383)
865 continue;
866 *pheads = heads;
867 *psectors = sectors;
868 *pcylinders = cylinders;
869 #if 0
870 printf("guessed geometry: LCHS=%d %d %d\n",
871 cylinders, heads, sectors);
872 #endif
873 return 0;
876 return -1;
879 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
881 int translation, lba_detected = 0;
882 int cylinders, heads, secs;
883 uint64_t nb_sectors;
885 /* if a geometry hint is available, use it */
886 bdrv_get_geometry(bs, &nb_sectors);
887 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
888 translation = bdrv_get_translation_hint(bs);
889 if (cylinders != 0) {
890 *pcyls = cylinders;
891 *pheads = heads;
892 *psecs = secs;
893 } else {
894 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
895 if (heads > 16) {
896 /* if heads > 16, it means that a BIOS LBA
897 translation was active, so the default
898 hardware geometry is OK */
899 lba_detected = 1;
900 goto default_geometry;
901 } else {
902 *pcyls = cylinders;
903 *pheads = heads;
904 *psecs = secs;
905 /* disable any translation to be in sync with
906 the logical geometry */
907 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
908 bdrv_set_translation_hint(bs,
909 BIOS_ATA_TRANSLATION_NONE);
912 } else {
913 default_geometry:
914 /* if no geometry, use a standard physical disk geometry */
915 cylinders = nb_sectors / (16 * 63);
917 if (cylinders > 16383)
918 cylinders = 16383;
919 else if (cylinders < 2)
920 cylinders = 2;
921 *pcyls = cylinders;
922 *pheads = 16;
923 *psecs = 63;
924 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
925 if ((*pcyls * *pheads) <= 131072) {
926 bdrv_set_translation_hint(bs,
927 BIOS_ATA_TRANSLATION_LARGE);
928 } else {
929 bdrv_set_translation_hint(bs,
930 BIOS_ATA_TRANSLATION_LBA);
934 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
938 void bdrv_set_geometry_hint(BlockDriverState *bs,
939 int cyls, int heads, int secs)
941 bs->cyls = cyls;
942 bs->heads = heads;
943 bs->secs = secs;
946 void bdrv_set_type_hint(BlockDriverState *bs, int type)
948 bs->type = type;
949 bs->removable = ((type == BDRV_TYPE_CDROM ||
950 type == BDRV_TYPE_FLOPPY));
953 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
955 bs->translation = translation;
958 void bdrv_get_geometry_hint(BlockDriverState *bs,
959 int *pcyls, int *pheads, int *psecs)
961 *pcyls = bs->cyls;
962 *pheads = bs->heads;
963 *psecs = bs->secs;
966 int bdrv_get_type_hint(BlockDriverState *bs)
968 return bs->type;
971 int bdrv_get_translation_hint(BlockDriverState *bs)
973 return bs->translation;
976 int bdrv_is_removable(BlockDriverState *bs)
978 return bs->removable;
981 int bdrv_is_read_only(BlockDriverState *bs)
983 return bs->read_only;
986 int bdrv_set_read_only(BlockDriverState *bs, int read_only)
988 int ret = bs->read_only;
989 bs->read_only = read_only;
990 return ret;
993 int bdrv_is_sg(BlockDriverState *bs)
995 return bs->sg;
998 int bdrv_enable_write_cache(BlockDriverState *bs)
1000 return bs->enable_write_cache;
1003 /* XXX: no longer used */
1004 void bdrv_set_change_cb(BlockDriverState *bs,
1005 void (*change_cb)(void *opaque), void *opaque)
1007 bs->change_cb = change_cb;
1008 bs->change_opaque = opaque;
1011 int bdrv_is_encrypted(BlockDriverState *bs)
1013 if (bs->backing_hd && bs->backing_hd->encrypted)
1014 return 1;
1015 return bs->encrypted;
1018 int bdrv_key_required(BlockDriverState *bs)
1020 BlockDriverState *backing_hd = bs->backing_hd;
1022 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1023 return 1;
1024 return (bs->encrypted && !bs->valid_key);
1027 int bdrv_set_key(BlockDriverState *bs, const char *key)
1029 int ret;
1030 if (bs->backing_hd && bs->backing_hd->encrypted) {
1031 ret = bdrv_set_key(bs->backing_hd, key);
1032 if (ret < 0)
1033 return ret;
1034 if (!bs->encrypted)
1035 return 0;
1037 if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
1038 return -1;
1039 ret = bs->drv->bdrv_set_key(bs, key);
1040 if (ret < 0) {
1041 bs->valid_key = 0;
1042 } else if (!bs->valid_key) {
1043 bs->valid_key = 1;
1044 /* call the change callback now, we skipped it on open */
1045 bs->media_changed = 1;
1046 if (bs->change_cb)
1047 bs->change_cb(bs->change_opaque);
1049 return ret;
1052 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1054 if (!bs->drv) {
1055 buf[0] = '\0';
1056 } else {
1057 pstrcpy(buf, buf_size, bs->drv->format_name);
1061 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1062 void *opaque)
1064 BlockDriver *drv;
1066 for (drv = first_drv; drv != NULL; drv = drv->next) {
1067 it(opaque, drv->format_name);
1071 BlockDriverState *bdrv_find(const char *name)
1073 BlockDriverState *bs;
1075 for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1076 if (!strcmp(name, bs->device_name))
1077 return bs;
1079 return NULL;
1082 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1084 BlockDriverState *bs;
1086 for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1087 it(opaque, bs);
1091 const char *bdrv_get_device_name(BlockDriverState *bs)
1093 return bs->device_name;
1096 void bdrv_flush(BlockDriverState *bs)
1098 if (!bs->drv)
1099 return;
1100 if (bs->drv->bdrv_flush)
1101 bs->drv->bdrv_flush(bs);
1102 if (bs->backing_hd)
1103 bdrv_flush(bs->backing_hd);
1106 void bdrv_flush_all(void)
1108 BlockDriverState *bs;
1110 for (bs = bdrv_first; bs != NULL; bs = bs->next)
1111 if (bs->drv && !bdrv_is_read_only(bs) &&
1112 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1113 bdrv_flush(bs);
1117 * Returns true iff the specified sector is present in the disk image. Drivers
1118 * not implementing the functionality are assumed to not support backing files,
1119 * hence all their sectors are reported as allocated.
1121 * 'pnum' is set to the number of sectors (including and immediately following
1122 * the specified sector) that are known to be in the same
1123 * allocated/unallocated state.
1125 * 'nb_sectors' is the max value 'pnum' should be set to.
1127 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1128 int *pnum)
1130 int64_t n;
1131 if (!bs->drv->bdrv_is_allocated) {
1132 if (sector_num >= bs->total_sectors) {
1133 *pnum = 0;
1134 return 0;
1136 n = bs->total_sectors - sector_num;
1137 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1138 return 1;
1140 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1143 static void bdrv_print_dict(QObject *obj, void *opaque)
1145 QDict *bs_dict;
1146 Monitor *mon = opaque;
1148 bs_dict = qobject_to_qdict(obj);
1150 monitor_printf(mon, "%s: type=%s removable=%d",
1151 qdict_get_str(bs_dict, "device"),
1152 qdict_get_str(bs_dict, "type"),
1153 qdict_get_bool(bs_dict, "removable"));
1155 if (qdict_get_bool(bs_dict, "removable")) {
1156 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1159 if (qdict_haskey(bs_dict, "inserted")) {
1160 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1162 monitor_printf(mon, " file=");
1163 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1164 if (qdict_haskey(qdict, "backing_file")) {
1165 monitor_printf(mon, " backing_file=");
1166 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1168 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1169 qdict_get_bool(qdict, "ro"),
1170 qdict_get_str(qdict, "drv"),
1171 qdict_get_bool(qdict, "encrypted"));
1172 } else {
1173 monitor_printf(mon, " [not inserted]");
1176 monitor_printf(mon, "\n");
1179 void bdrv_info_print(Monitor *mon, const QObject *data)
1181 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1185 * bdrv_info(): Block devices information
1187 * Each block device information is stored in a QDict and the
1188 * returned QObject is a QList of all devices.
1190 * The QDict contains the following:
1192 * - "device": device name
1193 * - "type": device type
1194 * - "removable": true if the device is removable, false otherwise
1195 * - "locked": true if the device is locked, false otherwise
1196 * - "inserted": only present if the device is inserted, it is a QDict
1197 * containing the following:
1198 * - "file": device file name
1199 * - "ro": true if read-only, false otherwise
1200 * - "drv": driver format name
1201 * - "backing_file": backing file name if one is used
1202 * - "encrypted": true if encrypted, false otherwise
1204 * Example:
1206 * [ { "device": "ide0-hd0", "type": "hd", "removable": false, "locked": false,
1207 * "inserted": { "file": "/tmp/foobar", "ro": false, "drv": "qcow2" } },
1208 * { "device": "floppy0", "type": "floppy", "removable": true,
1209 * "locked": false } ]
1211 void bdrv_info(Monitor *mon, QObject **ret_data)
1213 QList *bs_list;
1214 BlockDriverState *bs;
1216 bs_list = qlist_new();
1218 for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1219 QObject *bs_obj;
1220 const char *type = "unknown";
1222 switch(bs->type) {
1223 case BDRV_TYPE_HD:
1224 type = "hd";
1225 break;
1226 case BDRV_TYPE_CDROM:
1227 type = "cdrom";
1228 break;
1229 case BDRV_TYPE_FLOPPY:
1230 type = "floppy";
1231 break;
1234 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1235 "'removable': %i, 'locked': %i }",
1236 bs->device_name, type, bs->removable,
1237 bs->locked);
1238 assert(bs_obj != NULL);
1240 if (bs->drv) {
1241 QObject *obj;
1242 QDict *bs_dict = qobject_to_qdict(bs_obj);
1244 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1245 "'encrypted': %i }",
1246 bs->filename, bs->read_only,
1247 bs->drv->format_name,
1248 bdrv_is_encrypted(bs));
1249 assert(obj != NULL);
1250 if (bs->backing_file[0] != '\0') {
1251 QDict *qdict = qobject_to_qdict(obj);
1252 qdict_put(qdict, "backing_file",
1253 qstring_from_str(bs->backing_file));
1256 qdict_put_obj(bs_dict, "inserted", obj);
1258 qlist_append_obj(bs_list, bs_obj);
1261 *ret_data = QOBJECT(bs_list);
1264 static void bdrv_stats_iter(QObject *data, void *opaque)
1266 QDict *qdict;
1267 Monitor *mon = opaque;
1269 qdict = qobject_to_qdict(data);
1270 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1272 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1273 monitor_printf(mon, " rd_bytes=%" PRId64
1274 " wr_bytes=%" PRId64
1275 " rd_operations=%" PRId64
1276 " wr_operations=%" PRId64
1277 "\n",
1278 qdict_get_int(qdict, "rd_bytes"),
1279 qdict_get_int(qdict, "wr_bytes"),
1280 qdict_get_int(qdict, "rd_operations"),
1281 qdict_get_int(qdict, "wr_operations"));
1284 void bdrv_stats_print(Monitor *mon, const QObject *data)
1286 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1290 * bdrv_info_stats(): show block device statistics
1292 * Each device statistic information is stored in a QDict and
1293 * the returned QObject is a QList of all devices.
1295 * The QDict contains the following:
1297 * - "device": device name
1298 * - "stats": A QDict with the statistics information, it contains:
1299 * - "rd_bytes": bytes read
1300 * - "wr_bytes": bytes written
1301 * - "rd_operations": read operations
1302 * - "wr_operations": write operations
1304 * Example:
1306 * [ { "device": "ide0-hd0",
1307 * "stats": { "rd_bytes": 512,
1308 * "wr_bytes": 0,
1309 * "rd_operations": 1,
1310 * "wr_operations": 0 } },
1311 * { "device": "ide1-cd0",
1312 * "stats": { "rd_bytes": 0,
1313 * "wr_bytes": 0,
1314 * "rd_operations": 0,
1315 * "wr_operations": 0 } } ]
1317 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1319 QObject *obj;
1320 QList *devices;
1321 BlockDriverState *bs;
1323 devices = qlist_new();
1325 for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1326 obj = qobject_from_jsonf("{ 'device': %s, 'stats': {"
1327 "'rd_bytes': %" PRId64 ","
1328 "'wr_bytes': %" PRId64 ","
1329 "'rd_operations': %" PRId64 ","
1330 "'wr_operations': %" PRId64
1331 "} }",
1332 bs->device_name,
1333 bs->rd_bytes, bs->wr_bytes,
1334 bs->rd_ops, bs->wr_ops);
1335 assert(obj != NULL);
1336 qlist_append_obj(devices, obj);
1339 *ret_data = QOBJECT(devices);
1342 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1344 if (bs->backing_hd && bs->backing_hd->encrypted)
1345 return bs->backing_file;
1346 else if (bs->encrypted)
1347 return bs->filename;
1348 else
1349 return NULL;
1352 void bdrv_get_backing_filename(BlockDriverState *bs,
1353 char *filename, int filename_size)
1355 if (!bs->backing_hd) {
1356 pstrcpy(filename, filename_size, "");
1357 } else {
1358 pstrcpy(filename, filename_size, bs->backing_file);
1362 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1363 const uint8_t *buf, int nb_sectors)
1365 BlockDriver *drv = bs->drv;
1366 if (!drv)
1367 return -ENOMEDIUM;
1368 if (!drv->bdrv_write_compressed)
1369 return -ENOTSUP;
1370 if (bdrv_check_request(bs, sector_num, nb_sectors))
1371 return -EIO;
1373 if (bs->dirty_bitmap) {
1374 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1377 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1380 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1382 BlockDriver *drv = bs->drv;
1383 if (!drv)
1384 return -ENOMEDIUM;
1385 if (!drv->bdrv_get_info)
1386 return -ENOTSUP;
1387 memset(bdi, 0, sizeof(*bdi));
1388 return drv->bdrv_get_info(bs, bdi);
1391 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1392 int64_t pos, int size)
1394 BlockDriver *drv = bs->drv;
1395 if (!drv)
1396 return -ENOMEDIUM;
1397 if (!drv->bdrv_save_vmstate)
1398 return -ENOTSUP;
1399 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1402 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1403 int64_t pos, int size)
1405 BlockDriver *drv = bs->drv;
1406 if (!drv)
1407 return -ENOMEDIUM;
1408 if (!drv->bdrv_load_vmstate)
1409 return -ENOTSUP;
1410 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1413 /**************************************************************/
1414 /* handling of snapshots */
1416 int bdrv_snapshot_create(BlockDriverState *bs,
1417 QEMUSnapshotInfo *sn_info)
1419 BlockDriver *drv = bs->drv;
1420 if (!drv)
1421 return -ENOMEDIUM;
1422 if (!drv->bdrv_snapshot_create)
1423 return -ENOTSUP;
1424 return drv->bdrv_snapshot_create(bs, sn_info);
1427 int bdrv_snapshot_goto(BlockDriverState *bs,
1428 const char *snapshot_id)
1430 BlockDriver *drv = bs->drv;
1431 if (!drv)
1432 return -ENOMEDIUM;
1433 if (!drv->bdrv_snapshot_goto)
1434 return -ENOTSUP;
1435 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1438 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1440 BlockDriver *drv = bs->drv;
1441 if (!drv)
1442 return -ENOMEDIUM;
1443 if (!drv->bdrv_snapshot_delete)
1444 return -ENOTSUP;
1445 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1448 int bdrv_snapshot_list(BlockDriverState *bs,
1449 QEMUSnapshotInfo **psn_info)
1451 BlockDriver *drv = bs->drv;
1452 if (!drv)
1453 return -ENOMEDIUM;
1454 if (!drv->bdrv_snapshot_list)
1455 return -ENOTSUP;
1456 return drv->bdrv_snapshot_list(bs, psn_info);
1459 #define NB_SUFFIXES 4
1461 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1463 static const char suffixes[NB_SUFFIXES] = "KMGT";
1464 int64_t base;
1465 int i;
1467 if (size <= 999) {
1468 snprintf(buf, buf_size, "%" PRId64, size);
1469 } else {
1470 base = 1024;
1471 for(i = 0; i < NB_SUFFIXES; i++) {
1472 if (size < (10 * base)) {
1473 snprintf(buf, buf_size, "%0.1f%c",
1474 (double)size / base,
1475 suffixes[i]);
1476 break;
1477 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1478 snprintf(buf, buf_size, "%" PRId64 "%c",
1479 ((size + (base >> 1)) / base),
1480 suffixes[i]);
1481 break;
1483 base = base * 1024;
1486 return buf;
1489 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1491 char buf1[128], date_buf[128], clock_buf[128];
1492 #ifdef _WIN32
1493 struct tm *ptm;
1494 #else
1495 struct tm tm;
1496 #endif
1497 time_t ti;
1498 int64_t secs;
1500 if (!sn) {
1501 snprintf(buf, buf_size,
1502 "%-10s%-20s%7s%20s%15s",
1503 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1504 } else {
1505 ti = sn->date_sec;
1506 #ifdef _WIN32
1507 ptm = localtime(&ti);
1508 strftime(date_buf, sizeof(date_buf),
1509 "%Y-%m-%d %H:%M:%S", ptm);
1510 #else
1511 localtime_r(&ti, &tm);
1512 strftime(date_buf, sizeof(date_buf),
1513 "%Y-%m-%d %H:%M:%S", &tm);
1514 #endif
1515 secs = sn->vm_clock_nsec / 1000000000;
1516 snprintf(clock_buf, sizeof(clock_buf),
1517 "%02d:%02d:%02d.%03d",
1518 (int)(secs / 3600),
1519 (int)((secs / 60) % 60),
1520 (int)(secs % 60),
1521 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1522 snprintf(buf, buf_size,
1523 "%-10s%-20s%7s%20s%15s",
1524 sn->id_str, sn->name,
1525 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1526 date_buf,
1527 clock_buf);
1529 return buf;
1533 /**************************************************************/
1534 /* async I/Os */
1536 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1537 QEMUIOVector *qiov, int nb_sectors,
1538 BlockDriverCompletionFunc *cb, void *opaque)
1540 BlockDriver *drv = bs->drv;
1541 BlockDriverAIOCB *ret;
1543 if (!drv)
1544 return NULL;
1545 if (bdrv_check_request(bs, sector_num, nb_sectors))
1546 return NULL;
1548 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1549 cb, opaque);
1551 if (ret) {
1552 /* Update stats even though technically transfer has not happened. */
1553 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1554 bs->rd_ops ++;
1557 return ret;
1560 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1561 QEMUIOVector *qiov, int nb_sectors,
1562 BlockDriverCompletionFunc *cb, void *opaque)
1564 BlockDriver *drv = bs->drv;
1565 BlockDriverAIOCB *ret;
1567 if (!drv)
1568 return NULL;
1569 if (bs->read_only)
1570 return NULL;
1571 if (bdrv_check_request(bs, sector_num, nb_sectors))
1572 return NULL;
1574 if (bs->dirty_bitmap) {
1575 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1578 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1579 cb, opaque);
1581 if (ret) {
1582 /* Update stats even though technically transfer has not happened. */
1583 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1584 bs->wr_ops ++;
1587 return ret;
1591 typedef struct MultiwriteCB {
1592 int error;
1593 int num_requests;
1594 int num_callbacks;
1595 struct {
1596 BlockDriverCompletionFunc *cb;
1597 void *opaque;
1598 QEMUIOVector *free_qiov;
1599 void *free_buf;
1600 } callbacks[];
1601 } MultiwriteCB;
1603 static void multiwrite_user_cb(MultiwriteCB *mcb)
1605 int i;
1607 for (i = 0; i < mcb->num_callbacks; i++) {
1608 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1609 qemu_free(mcb->callbacks[i].free_qiov);
1610 qemu_free(mcb->callbacks[i].free_buf);
1614 static void multiwrite_cb(void *opaque, int ret)
1616 MultiwriteCB *mcb = opaque;
1618 if (ret < 0) {
1619 mcb->error = ret;
1620 multiwrite_user_cb(mcb);
1623 mcb->num_requests--;
1624 if (mcb->num_requests == 0) {
1625 if (mcb->error == 0) {
1626 multiwrite_user_cb(mcb);
1628 qemu_free(mcb);
1632 static int multiwrite_req_compare(const void *a, const void *b)
1634 return (((BlockRequest*) a)->sector - ((BlockRequest*) b)->sector);
1638 * Takes a bunch of requests and tries to merge them. Returns the number of
1639 * requests that remain after merging.
1641 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1642 int num_reqs, MultiwriteCB *mcb)
1644 int i, outidx;
1646 // Sort requests by start sector
1647 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1649 // Check if adjacent requests touch the same clusters. If so, combine them,
1650 // filling up gaps with zero sectors.
1651 outidx = 0;
1652 for (i = 1; i < num_reqs; i++) {
1653 int merge = 0;
1654 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1656 // This handles the cases that are valid for all block drivers, namely
1657 // exactly sequential writes and overlapping writes.
1658 if (reqs[i].sector <= oldreq_last) {
1659 merge = 1;
1662 // The block driver may decide that it makes sense to combine requests
1663 // even if there is a gap of some sectors between them. In this case,
1664 // the gap is filled with zeros (therefore only applicable for yet
1665 // unused space in format like qcow2).
1666 if (!merge && bs->drv->bdrv_merge_requests) {
1667 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
1670 if (merge) {
1671 size_t size;
1672 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
1673 qemu_iovec_init(qiov,
1674 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1676 // Add the first request to the merged one. If the requests are
1677 // overlapping, drop the last sectors of the first request.
1678 size = (reqs[i].sector - reqs[outidx].sector) << 9;
1679 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
1681 // We might need to add some zeros between the two requests
1682 if (reqs[i].sector > oldreq_last) {
1683 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
1684 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
1685 memset(buf, 0, zero_bytes);
1686 qemu_iovec_add(qiov, buf, zero_bytes);
1687 mcb->callbacks[i].free_buf = buf;
1690 // Add the second request
1691 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
1693 reqs[outidx].nb_sectors += reqs[i].nb_sectors;
1694 reqs[outidx].qiov = qiov;
1696 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1697 } else {
1698 outidx++;
1699 reqs[outidx].sector = reqs[i].sector;
1700 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1701 reqs[outidx].qiov = reqs[i].qiov;
1705 return outidx + 1;
1709 * Submit multiple AIO write requests at once.
1711 * On success, the function returns 0 and all requests in the reqs array have
1712 * been submitted. In error case this function returns -1, and any of the
1713 * requests may or may not be submitted yet. In particular, this means that the
1714 * callback will be called for some of the requests, for others it won't. The
1715 * caller must check the error field of the BlockRequest to wait for the right
1716 * callbacks (if error != 0, no callback will be called).
1718 * The implementation may modify the contents of the reqs array, e.g. to merge
1719 * requests. However, the fields opaque and error are left unmodified as they
1720 * are used to signal failure for a single request to the caller.
1722 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1724 BlockDriverAIOCB *acb;
1725 MultiwriteCB *mcb;
1726 int i;
1728 if (num_reqs == 0) {
1729 return 0;
1732 // Create MultiwriteCB structure
1733 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1734 mcb->num_requests = 0;
1735 mcb->num_callbacks = num_reqs;
1737 for (i = 0; i < num_reqs; i++) {
1738 mcb->callbacks[i].cb = reqs[i].cb;
1739 mcb->callbacks[i].opaque = reqs[i].opaque;
1742 // Check for mergable requests
1743 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1745 // Run the aio requests
1746 for (i = 0; i < num_reqs; i++) {
1747 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
1748 reqs[i].nb_sectors, multiwrite_cb, mcb);
1750 if (acb == NULL) {
1751 // We can only fail the whole thing if no request has been
1752 // submitted yet. Otherwise we'll wait for the submitted AIOs to
1753 // complete and report the error in the callback.
1754 if (mcb->num_requests == 0) {
1755 reqs[i].error = EIO;
1756 goto fail;
1757 } else {
1758 mcb->error = EIO;
1759 break;
1761 } else {
1762 mcb->num_requests++;
1766 return 0;
1768 fail:
1769 free(mcb);
1770 return -1;
1773 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
1774 BlockDriverCompletionFunc *cb, void *opaque)
1776 BlockDriver *drv = bs->drv;
1778 if (!drv)
1779 return NULL;
1782 * Note that unlike bdrv_flush the driver is reponsible for flushing a
1783 * backing image if it exists.
1785 return drv->bdrv_aio_flush(bs, cb, opaque);
1788 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1790 acb->pool->cancel(acb);
1794 /**************************************************************/
1795 /* async block device emulation */
1797 typedef struct BlockDriverAIOCBSync {
1798 BlockDriverAIOCB common;
1799 QEMUBH *bh;
1800 int ret;
1801 /* vector translation state */
1802 QEMUIOVector *qiov;
1803 uint8_t *bounce;
1804 int is_write;
1805 } BlockDriverAIOCBSync;
1807 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1809 BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1810 qemu_bh_delete(acb->bh);
1811 acb->bh = NULL;
1812 qemu_aio_release(acb);
1815 static AIOPool bdrv_em_aio_pool = {
1816 .aiocb_size = sizeof(BlockDriverAIOCBSync),
1817 .cancel = bdrv_aio_cancel_em,
1820 static void bdrv_aio_bh_cb(void *opaque)
1822 BlockDriverAIOCBSync *acb = opaque;
1824 if (!acb->is_write)
1825 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1826 qemu_vfree(acb->bounce);
1827 acb->common.cb(acb->common.opaque, acb->ret);
1828 qemu_bh_delete(acb->bh);
1829 acb->bh = NULL;
1830 qemu_aio_release(acb);
1833 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1834 int64_t sector_num,
1835 QEMUIOVector *qiov,
1836 int nb_sectors,
1837 BlockDriverCompletionFunc *cb,
1838 void *opaque,
1839 int is_write)
1842 BlockDriverAIOCBSync *acb;
1844 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1845 acb->is_write = is_write;
1846 acb->qiov = qiov;
1847 acb->bounce = qemu_blockalign(bs, qiov->size);
1849 if (!acb->bh)
1850 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1852 if (is_write) {
1853 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1854 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1855 } else {
1856 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1859 qemu_bh_schedule(acb->bh);
1861 return &acb->common;
1864 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1865 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1866 BlockDriverCompletionFunc *cb, void *opaque)
1868 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1871 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1872 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1873 BlockDriverCompletionFunc *cb, void *opaque)
1875 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1878 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
1879 BlockDriverCompletionFunc *cb, void *opaque)
1881 BlockDriverAIOCBSync *acb;
1883 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1884 acb->is_write = 1; /* don't bounce in the completion hadler */
1885 acb->qiov = NULL;
1886 acb->bounce = NULL;
1887 acb->ret = 0;
1889 if (!acb->bh)
1890 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1892 bdrv_flush(bs);
1893 qemu_bh_schedule(acb->bh);
1894 return &acb->common;
1897 /**************************************************************/
1898 /* sync block device emulation */
1900 static void bdrv_rw_em_cb(void *opaque, int ret)
1902 *(int *)opaque = ret;
1905 #define NOT_DONE 0x7fffffff
1907 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1908 uint8_t *buf, int nb_sectors)
1910 int async_ret;
1911 BlockDriverAIOCB *acb;
1912 struct iovec iov;
1913 QEMUIOVector qiov;
1915 async_context_push();
1917 async_ret = NOT_DONE;
1918 iov.iov_base = (void *)buf;
1919 iov.iov_len = nb_sectors * 512;
1920 qemu_iovec_init_external(&qiov, &iov, 1);
1921 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1922 bdrv_rw_em_cb, &async_ret);
1923 if (acb == NULL) {
1924 async_ret = -1;
1925 goto fail;
1928 while (async_ret == NOT_DONE) {
1929 qemu_aio_wait();
1933 fail:
1934 async_context_pop();
1935 return async_ret;
1938 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1939 const uint8_t *buf, int nb_sectors)
1941 int async_ret;
1942 BlockDriverAIOCB *acb;
1943 struct iovec iov;
1944 QEMUIOVector qiov;
1946 async_context_push();
1948 async_ret = NOT_DONE;
1949 iov.iov_base = (void *)buf;
1950 iov.iov_len = nb_sectors * 512;
1951 qemu_iovec_init_external(&qiov, &iov, 1);
1952 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1953 bdrv_rw_em_cb, &async_ret);
1954 if (acb == NULL) {
1955 async_ret = -1;
1956 goto fail;
1958 while (async_ret == NOT_DONE) {
1959 qemu_aio_wait();
1962 fail:
1963 async_context_pop();
1964 return async_ret;
1967 void bdrv_init(void)
1969 module_call_init(MODULE_INIT_BLOCK);
1972 void bdrv_init_with_whitelist(void)
1974 use_bdrv_whitelist = 1;
1975 bdrv_init();
1978 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
1979 BlockDriverCompletionFunc *cb, void *opaque)
1981 BlockDriverAIOCB *acb;
1983 if (pool->free_aiocb) {
1984 acb = pool->free_aiocb;
1985 pool->free_aiocb = acb->next;
1986 } else {
1987 acb = qemu_mallocz(pool->aiocb_size);
1988 acb->pool = pool;
1990 acb->bs = bs;
1991 acb->cb = cb;
1992 acb->opaque = opaque;
1993 return acb;
1996 void qemu_aio_release(void *p)
1998 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
1999 AIOPool *pool = acb->pool;
2000 acb->next = pool->free_aiocb;
2001 pool->free_aiocb = acb;
2004 /**************************************************************/
2005 /* removable device support */
2008 * Return TRUE if the media is present
2010 int bdrv_is_inserted(BlockDriverState *bs)
2012 BlockDriver *drv = bs->drv;
2013 int ret;
2014 if (!drv)
2015 return 0;
2016 if (!drv->bdrv_is_inserted)
2017 return 1;
2018 ret = drv->bdrv_is_inserted(bs);
2019 return ret;
2023 * Return TRUE if the media changed since the last call to this
2024 * function. It is currently only used for floppy disks
2026 int bdrv_media_changed(BlockDriverState *bs)
2028 BlockDriver *drv = bs->drv;
2029 int ret;
2031 if (!drv || !drv->bdrv_media_changed)
2032 ret = -ENOTSUP;
2033 else
2034 ret = drv->bdrv_media_changed(bs);
2035 if (ret == -ENOTSUP)
2036 ret = bs->media_changed;
2037 bs->media_changed = 0;
2038 return ret;
2042 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2044 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2046 BlockDriver *drv = bs->drv;
2047 int ret;
2049 if (bs->locked) {
2050 return -EBUSY;
2053 if (!drv || !drv->bdrv_eject) {
2054 ret = -ENOTSUP;
2055 } else {
2056 ret = drv->bdrv_eject(bs, eject_flag);
2058 if (ret == -ENOTSUP) {
2059 if (eject_flag)
2060 bdrv_close(bs);
2061 ret = 0;
2064 return ret;
2067 int bdrv_is_locked(BlockDriverState *bs)
2069 return bs->locked;
2073 * Lock or unlock the media (if it is locked, the user won't be able
2074 * to eject it manually).
2076 void bdrv_set_locked(BlockDriverState *bs, int locked)
2078 BlockDriver *drv = bs->drv;
2080 bs->locked = locked;
2081 if (drv && drv->bdrv_set_locked) {
2082 drv->bdrv_set_locked(bs, locked);
2086 /* needed for generic scsi interface */
2088 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2090 BlockDriver *drv = bs->drv;
2092 if (drv && drv->bdrv_ioctl)
2093 return drv->bdrv_ioctl(bs, req, buf);
2094 return -ENOTSUP;
2097 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2098 unsigned long int req, void *buf,
2099 BlockDriverCompletionFunc *cb, void *opaque)
2101 BlockDriver *drv = bs->drv;
2103 if (drv && drv->bdrv_aio_ioctl)
2104 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2105 return NULL;
2110 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2112 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2115 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2117 int64_t bitmap_size;
2119 if (enable) {
2120 if (!bs->dirty_bitmap) {
2121 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2122 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2123 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2125 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2127 } else {
2128 if (bs->dirty_bitmap) {
2129 qemu_free(bs->dirty_bitmap);
2130 bs->dirty_bitmap = NULL;
2135 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2137 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2139 if (bs->dirty_bitmap &&
2140 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2141 return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2142 (1 << (chunk % (sizeof(unsigned long) * 8)));
2143 } else {
2144 return 0;
2148 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2149 int nr_sectors)
2151 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);