?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Use qemu_mallocz() instead of calloc() in img_convert()
[qemu.git] / block.c
blobe7a986c48a111cfeabde278cff29be806897c44a
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
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"
31
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
41
42 #ifdef _WIN32
43 #include <windows.h>
44 #endif
45
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);
60
61 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
62 QTAILQ_HEAD_INITIALIZER(bdrv_states);
63
64 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
65 QLIST_HEAD_INITIALIZER(bdrv_drivers);
66
67 /* The device to use for VM snapshots */
68 static BlockDriverState *bs_snapshots;
69
70 /* If non-zero, use only whitelisted block drivers */
71 static int use_bdrv_whitelist;
72
73 int path_is_absolute(const char *path)
74 {
75 const char *p;
76 #ifdef _WIN32
77 /* specific case for names like: "\\.\d:" */
78 if (*path == '/' || *path == '\\')
79 return 1;
80 #endif
81 p = strchr(path, ':');
82 if (p)
83 p++;
84 else
85 p = path;
86 #ifdef _WIN32
87 return (*p == '/' || *p == '\\');
88 #else
89 return (*p == '/');
90 #endif
91 }
92
93 /* if filename is absolute, just copy it to dest. Otherwise, build a
94 path to it by considering it is relative to base_path. URL are
95 supported. */
96 void path_combine(char *dest, int dest_size,
97 const char *base_path,
98 const char *filename)
99 {
100 const char *p, *p1;
101 int len;
102
103 if (dest_size <= 0)
104 return;
105 if (path_is_absolute(filename)) {
106 pstrcpy(dest, dest_size, filename);
107 } else {
108 p = strchr(base_path, ':');
109 if (p)
110 p++;
111 else
112 p = base_path;
113 p1 = strrchr(base_path, '/');
114 #ifdef _WIN32
115 {
116 const char *p2;
117 p2 = strrchr(base_path, '\\');
118 if (!p1 || p2 > p1)
119 p1 = p2;
120 }
121 #endif
122 if (p1)
123 p1++;
124 else
125 p1 = base_path;
126 if (p1 > p)
127 p = p1;
128 len = p - base_path;
129 if (len > dest_size - 1)
130 len = dest_size - 1;
131 memcpy(dest, base_path, len);
132 dest[len] = '\0';
133 pstrcat(dest, dest_size, filename);
134 }
135 }
136
137 void bdrv_register(BlockDriver *bdrv)
138 {
139 if (!bdrv->bdrv_aio_readv) {
140 /* add AIO emulation layer */
141 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
142 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
143 } else if (!bdrv->bdrv_read) {
144 /* add synchronous IO emulation layer */
145 bdrv->bdrv_read = bdrv_read_em;
146 bdrv->bdrv_write = bdrv_write_em;
147 }
148
149 if (!bdrv->bdrv_aio_flush)
150 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
151
152 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
153 }
154
155 /* create a new block device (by default it is empty) */
156 BlockDriverState *bdrv_new(const char *device_name)
157 {
158 BlockDriverState *bs;
159
160 bs = qemu_mallocz(sizeof(BlockDriverState));
161 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
162 if (device_name[0] != '\0') {
163 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
164 }
165 return bs;
166 }
167
168 BlockDriver *bdrv_find_format(const char *format_name)
169 {
170 BlockDriver *drv1;
171 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
172 if (!strcmp(drv1->format_name, format_name)) {
173 return drv1;
174 }
175 }
176 return NULL;
177 }
178
179 static int bdrv_is_whitelisted(BlockDriver *drv)
180 {
181 static const char *whitelist[] = {
182 CONFIG_BDRV_WHITELIST
183 };
184 const char **p;
185
186 if (!whitelist[0])
187 return 1; /* no whitelist, anything goes */
188
189 for (p = whitelist; *p; p++) {
190 if (!strcmp(drv->format_name, *p)) {
191 return 1;
192 }
193 }
194 return 0;
195 }
196
197 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
198 {
199 BlockDriver *drv = bdrv_find_format(format_name);
200 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
201 }
202
203 int bdrv_create(BlockDriver *drv, const char* filename,
204 QEMUOptionParameter *options)
205 {
206 if (!drv->bdrv_create)
207 return -ENOTSUP;
208
209 return drv->bdrv_create(filename, options);
210 }
211
212 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
213 {
214 BlockDriver *drv;
215
216 drv = bdrv_find_protocol(filename);
217 if (drv == NULL) {
218 return -ENOENT;
219 }
220
221 return bdrv_create(drv, filename, options);
222 }
223
224 #ifdef _WIN32
225 void get_tmp_filename(char *filename, int size)
226 {
227 char temp_dir[MAX_PATH];
228
229 GetTempPath(MAX_PATH, temp_dir);
230 GetTempFileName(temp_dir, "qem", 0, filename);
231 }
232 #else
233 void get_tmp_filename(char *filename, int size)
234 {
235 int fd;
236 const char *tmpdir;
237 /* XXX: race condition possible */
238 tmpdir = getenv("TMPDIR");
239 if (!tmpdir)
240 tmpdir = "/tmp";
241 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
242 fd = mkstemp(filename);
243 close(fd);
244 }
245 #endif
246
247 #ifdef _WIN32
248 static int is_windows_drive_prefix(const char *filename)
249 {
250 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
251 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
252 filename[1] == ':');
253 }
254
255 int is_windows_drive(const char *filename)
256 {
257 if (is_windows_drive_prefix(filename) &&
258 filename[2] == '\0')
259 return 1;
260 if (strstart(filename, "\\\\.\\", NULL) ||
261 strstart(filename, "//./", NULL))
262 return 1;
263 return 0;
264 }
265 #endif
266
267 /*
268 * Detect host devices. By convention, /dev/cdrom[N] is always
269 * recognized as a host CDROM.
270 */
271 static BlockDriver *find_hdev_driver(const char *filename)
272 {
273 int score_max = 0, score;
274 BlockDriver *drv = NULL, *d;
275
276 QLIST_FOREACH(d, &bdrv_drivers, list) {
277 if (d->bdrv_probe_device) {
278 score = d->bdrv_probe_device(filename);
279 if (score > score_max) {
280 score_max = score;
281 drv = d;
282 }
283 }
284 }
285
286 return drv;
287 }
288
289 BlockDriver *bdrv_find_protocol(const char *filename)
290 {
291 BlockDriver *drv1;
292 char protocol[128];
293 int len;
294 const char *p;
295
296 /* TODO Drivers without bdrv_file_open must be specified explicitly */
297
298 /*
299 * XXX(hch): we really should not let host device detection
300 * override an explicit protocol specification, but moving this
301 * later breaks access to device names with colons in them.
302 * Thanks to the brain-dead persistent naming schemes on udev-
303 * based Linux systems those actually are quite common.
304 */
305 drv1 = find_hdev_driver(filename);
306 if (drv1) {
307 return drv1;
308 }
309
310 #ifdef _WIN32
311 if (is_windows_drive(filename) ||
312 is_windows_drive_prefix(filename))
313 return bdrv_find_format("file");
314 #endif
315
316 p = strchr(filename, ':');
317 if (!p) {
318 return bdrv_find_format("file");
319 }
320 len = p - filename;
321 if (len > sizeof(protocol) - 1)
322 len = sizeof(protocol) - 1;
323 memcpy(protocol, filename, len);
324 protocol[len] = '\0';
325 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
326 if (drv1->protocol_name &&
327 !strcmp(drv1->protocol_name, protocol)) {
328 return drv1;
329 }
330 }
331 return NULL;
332 }
333
334 static int find_image_format(const char *filename, BlockDriver **pdrv)
335 {
336 int ret, score, score_max;
337 BlockDriver *drv1, *drv;
338 uint8_t buf[2048];
339 BlockDriverState *bs;
340
341 ret = bdrv_file_open(&bs, filename, 0);
342 if (ret < 0) {
343 *pdrv = NULL;
344 return ret;
345 }
346
347 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
348 if (bs->sg || !bdrv_is_inserted(bs)) {
349 bdrv_delete(bs);
350 drv = bdrv_find_format("raw");
351 if (!drv) {
352 ret = -ENOENT;
353 }
354 *pdrv = drv;
355 return ret;
356 }
357
358 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
359 bdrv_delete(bs);
360 if (ret < 0) {
361 *pdrv = NULL;
362 return ret;
363 }
364
365 score_max = 0;
366 drv = NULL;
367 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
368 if (drv1->bdrv_probe) {
369 score = drv1->bdrv_probe(buf, ret, filename);
370 if (score > score_max) {
371 score_max = score;
372 drv = drv1;
373 }
374 }
375 }
376 if (!drv) {
377 ret = -ENOENT;
378 }
379 *pdrv = drv;
380 return ret;
381 }
382
383 /**
384 * Set the current 'total_sectors' value
385 */
386 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
387 {
388 BlockDriver *drv = bs->drv;
389
390 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
391 if (bs->sg)
392 return 0;
393
394 /* query actual device if possible, otherwise just trust the hint */
395 if (drv->bdrv_getlength) {
396 int64_t length = drv->bdrv_getlength(bs);
397 if (length < 0) {
398 return length;
399 }
400 hint = length >> BDRV_SECTOR_BITS;
401 }
402
403 bs->total_sectors = hint;
404 return 0;
405 }
406
407 /*
408 * Common part for opening disk images and files
409 */
410 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
411 int flags, BlockDriver *drv)
412 {
413 int ret, open_flags;
414
415 assert(drv != NULL);
416
417 bs->file = NULL;
418 bs->total_sectors = 0;
419 bs->encrypted = 0;
420 bs->valid_key = 0;
421 bs->open_flags = flags;
422 /* buffer_alignment defaulted to 512, drivers can change this value */
423 bs->buffer_alignment = 512;
424
425 pstrcpy(bs->filename, sizeof(bs->filename), filename);
426
427 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
428 return -ENOTSUP;
429 }
430
431 bs->drv = drv;
432 bs->opaque = qemu_mallocz(drv->instance_size);
433
434 /*
435 * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
436 * write cache to the guest. We do need the fdatasync to flush
437 * out transactions for block allocations, and we maybe have a
438 * volatile write cache in our backing device to deal with.
439 */
440 if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
441 bs->enable_write_cache = 1;
442
443 /*
444 * Clear flags that are internal to the block layer before opening the
445 * image.
446 */
447 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
448
449 /*
450 * Snapshots should be writeable.
451 */
452 if (bs->is_temporary) {
453 open_flags |= BDRV_O_RDWR;
454 }
455
456 /* Open the image, either directly or using a protocol */
457 if (drv->bdrv_file_open) {
458 ret = drv->bdrv_file_open(bs, filename, open_flags);
459 } else {
460 ret = bdrv_file_open(&bs->file, filename, open_flags);
461 if (ret >= 0) {
462 ret = drv->bdrv_open(bs, open_flags);
463 }
464 }
465
466 if (ret < 0) {
467 goto free_and_fail;
468 }
469
470 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
471
472 ret = refresh_total_sectors(bs, bs->total_sectors);
473 if (ret < 0) {
474 goto free_and_fail;
475 }
476
477 #ifndef _WIN32
478 if (bs->is_temporary) {
479 unlink(filename);
480 }
481 #endif
482 return 0;
483
484 free_and_fail:
485 if (bs->file) {
486 bdrv_delete(bs->file);
487 bs->file = NULL;
488 }
489 qemu_free(bs->opaque);
490 bs->opaque = NULL;
491 bs->drv = NULL;
492 return ret;
493 }
494
495 /*
496 * Opens a file using a protocol (file, host_device, nbd, ...)
497 */
498 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
499 {
500 BlockDriverState *bs;
501 BlockDriver *drv;
502 int ret;
503
504 drv = bdrv_find_protocol(filename);
505 if (!drv) {
506 return -ENOENT;
507 }
508
509 bs = bdrv_new("");
510 ret = bdrv_open_common(bs, filename, flags, drv);
511 if (ret < 0) {
512 bdrv_delete(bs);
513 return ret;
514 }
515 bs->growable = 1;
516 *pbs = bs;
517 return 0;
518 }
519
520 /*
521 * Opens a disk image (raw, qcow2, vmdk, ...)
522 */
523 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
524 BlockDriver *drv)
525 {
526 int ret;
527
528 if (flags & BDRV_O_SNAPSHOT) {
529 BlockDriverState *bs1;
530 int64_t total_size;
531 int is_protocol = 0;
532 BlockDriver *bdrv_qcow2;
533 QEMUOptionParameter *options;
534 char tmp_filename[PATH_MAX];
535 char backing_filename[PATH_MAX];
536
537 /* if snapshot, we create a temporary backing file and open it
538 instead of opening 'filename' directly */
539
540 /* if there is a backing file, use it */
541 bs1 = bdrv_new("");
542 ret = bdrv_open(bs1, filename, 0, drv);
543 if (ret < 0) {
544 bdrv_delete(bs1);
545 return ret;
546 }
547 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
548
549 if (bs1->drv && bs1->drv->protocol_name)
550 is_protocol = 1;
551
552 bdrv_delete(bs1);
553
554 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
555
556 /* Real path is meaningless for protocols */
557 if (is_protocol)
558 snprintf(backing_filename, sizeof(backing_filename),
559 "%s", filename);
560 else if (!realpath(filename, backing_filename))
561 return -errno;
562
563 bdrv_qcow2 = bdrv_find_format("qcow2");
564 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
565
566 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
567 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
568 if (drv) {
569 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
570 drv->format_name);
571 }
572
573 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
574 free_option_parameters(options);
575 if (ret < 0) {
576 return ret;
577 }
578
579 filename = tmp_filename;
580 drv = bdrv_qcow2;
581 bs->is_temporary = 1;
582 }
583
584 /* Find the right image format driver */
585 if (!drv) {
586 ret = find_image_format(filename, &drv);
587 }
588
589 if (!drv) {
590 goto unlink_and_fail;
591 }
592
593 /* Open the image */
594 ret = bdrv_open_common(bs, filename, flags, drv);
595 if (ret < 0) {
596 goto unlink_and_fail;
597 }
598
599 /* If there is a backing file, use it */
600 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
601 char backing_filename[PATH_MAX];
602 int back_flags;
603 BlockDriver *back_drv = NULL;
604
605 bs->backing_hd = bdrv_new("");
606 path_combine(backing_filename, sizeof(backing_filename),
607 filename, bs->backing_file);
608 if (bs->backing_format[0] != '\0')
609 back_drv = bdrv_find_format(bs->backing_format);
610
611 /* backing files always opened read-only */
612 back_flags =
613 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
614
615 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
616 if (ret < 0) {
617 bdrv_close(bs);
618 return ret;
619 }
620 if (bs->is_temporary) {
621 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
622 } else {
623 /* base image inherits from "parent" */
624 bs->backing_hd->keep_read_only = bs->keep_read_only;
625 }
626 }
627
628 if (!bdrv_key_required(bs)) {
629 /* call the change callback */
630 bs->media_changed = 1;
631 if (bs->change_cb)
632 bs->change_cb(bs->change_opaque);
633 }
634
635 return 0;
636
637 unlink_and_fail:
638 if (bs->is_temporary) {
639 unlink(filename);
640 }
641 return ret;
642 }
643
644 void bdrv_close(BlockDriverState *bs)
645 {
646 if (bs->drv) {
647 if (bs == bs_snapshots) {
648 bs_snapshots = NULL;
649 }
650 if (bs->backing_hd) {
651 bdrv_delete(bs->backing_hd);
652 bs->backing_hd = NULL;
653 }
654 bs->drv->bdrv_close(bs);
655 qemu_free(bs->opaque);
656 #ifdef _WIN32
657 if (bs->is_temporary) {
658 unlink(bs->filename);
659 }
660 #endif
661 bs->opaque = NULL;
662 bs->drv = NULL;
663
664 if (bs->file != NULL) {
665 bdrv_close(bs->file);
666 }
667
668 /* call the change callback */
669 bs->media_changed = 1;
670 if (bs->change_cb)
671 bs->change_cb(bs->change_opaque);
672 }
673 }
674
675 void bdrv_close_all(void)
676 {
677 BlockDriverState *bs;
678
679 QTAILQ_FOREACH(bs, &bdrv_states, list) {
680 bdrv_close(bs);
681 }
682 }
683
684 void bdrv_delete(BlockDriverState *bs)
685 {
686 assert(!bs->peer);
687
688 /* remove from list, if necessary */
689 if (bs->device_name[0] != '\0') {
690 QTAILQ_REMOVE(&bdrv_states, bs, list);
691 }
692
693 bdrv_close(bs);
694 if (bs->file != NULL) {
695 bdrv_delete(bs->file);
696 }
697
698 assert(bs != bs_snapshots);
699 qemu_free(bs);
700 }
701
702 int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
703 {
704 if (bs->peer) {
705 return -EBUSY;
706 }
707 bs->peer = qdev;
708 return 0;
709 }
710
711 void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
712 {
713 assert(bs->peer == qdev);
714 bs->peer = NULL;
715 }
716
717 DeviceState *bdrv_get_attached(BlockDriverState *bs)
718 {
719 return bs->peer;
720 }
721
722 /*
723 * Run consistency checks on an image
724 *
725 * Returns 0 if the check could be completed (it doesn't mean that the image is
726 * free of errors) or -errno when an internal error occured. The results of the
727 * check are stored in res.
728 */
729 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
730 {
731 if (bs->drv->bdrv_check == NULL) {
732 return -ENOTSUP;
733 }
734
735 memset(res, 0, sizeof(*res));
736 return bs->drv->bdrv_check(bs, res);
737 }
738
739 #define COMMIT_BUF_SECTORS 2048
740
741 /* commit COW file into the raw image */
742 int bdrv_commit(BlockDriverState *bs)
743 {
744 BlockDriver *drv = bs->drv;
745 BlockDriver *backing_drv;
746 int64_t sector, total_sectors;
747 int n, ro, open_flags;
748 int ret = 0, rw_ret = 0;
749 uint8_t *buf;
750 char filename[1024];
751 BlockDriverState *bs_rw, *bs_ro;
752
753 if (!drv)
754 return -ENOMEDIUM;
755
756 if (!bs->backing_hd) {
757 return -ENOTSUP;
758 }
759
760 if (bs->backing_hd->keep_read_only) {
761 return -EACCES;
762 }
763
764 backing_drv = bs->backing_hd->drv;
765 ro = bs->backing_hd->read_only;
766 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
767 open_flags = bs->backing_hd->open_flags;
768
769 if (ro) {
770 /* re-open as RW */
771 bdrv_delete(bs->backing_hd);
772 bs->backing_hd = NULL;
773 bs_rw = bdrv_new("");
774 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
775 backing_drv);
776 if (rw_ret < 0) {
777 bdrv_delete(bs_rw);
778 /* try to re-open read-only */
779 bs_ro = bdrv_new("");
780 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
781 backing_drv);
782 if (ret < 0) {
783 bdrv_delete(bs_ro);
784 /* drive not functional anymore */
785 bs->drv = NULL;
786 return ret;
787 }
788 bs->backing_hd = bs_ro;
789 return rw_ret;
790 }
791 bs->backing_hd = bs_rw;
792 }
793
794 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
795 buf = qemu_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
796
797 for (sector = 0; sector < total_sectors; sector += n) {
798 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
799
800 if (bdrv_read(bs, sector, buf, n) != 0) {
801 ret = -EIO;
802 goto ro_cleanup;
803 }
804
805 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
806 ret = -EIO;
807 goto ro_cleanup;
808 }
809 }
810 }
811
812 if (drv->bdrv_make_empty) {
813 ret = drv->bdrv_make_empty(bs);
814 bdrv_flush(bs);
815 }
816
817 /*
818 * Make sure all data we wrote to the backing device is actually
819 * stable on disk.
820 */
821 if (bs->backing_hd)
822 bdrv_flush(bs->backing_hd);
823
824 ro_cleanup:
825 qemu_free(buf);
826
827 if (ro) {
828 /* re-open as RO */
829 bdrv_delete(bs->backing_hd);
830 bs->backing_hd = NULL;
831 bs_ro = bdrv_new("");
832 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
833 backing_drv);
834 if (ret < 0) {
835 bdrv_delete(bs_ro);
836 /* drive not functional anymore */
837 bs->drv = NULL;
838 return ret;
839 }
840 bs->backing_hd = bs_ro;
841 bs->backing_hd->keep_read_only = 0;
842 }
843
844 return ret;
845 }
846
847 void bdrv_commit_all(void)
848 {
849 BlockDriverState *bs;
850
851 QTAILQ_FOREACH(bs, &bdrv_states, list) {
852 bdrv_commit(bs);
853 }
854 }
855
856 /*
857 * Return values:
858 * 0 - success
859 * -EINVAL - backing format specified, but no file
860 * -ENOSPC - can't update the backing file because no space is left in the
861 * image file header
862 * -ENOTSUP - format driver doesn't support changing the backing file
863 */
864 int bdrv_change_backing_file(BlockDriverState *bs,
865 const char *backing_file, const char *backing_fmt)
866 {
867 BlockDriver *drv = bs->drv;
868
869 if (drv->bdrv_change_backing_file != NULL) {
870 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
871 } else {
872 return -ENOTSUP;
873 }
874 }
875
876 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
877 size_t size)
878 {
879 int64_t len;
880
881 if (!bdrv_is_inserted(bs))
882 return -ENOMEDIUM;
883
884 if (bs->growable)
885 return 0;
886
887 len = bdrv_getlength(bs);
888
889 if (offset < 0)
890 return -EIO;
891
892 if ((offset > len) || (len - offset < size))
893 return -EIO;
894
895 return 0;
896 }
897
898 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
899 int nb_sectors)
900 {
901 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
902 nb_sectors * BDRV_SECTOR_SIZE);
903 }
904
905 /* return < 0 if error. See bdrv_write() for the return codes */
906 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
907 uint8_t *buf, int nb_sectors)
908 {
909 BlockDriver *drv = bs->drv;
910
911 if (!drv)
912 return -ENOMEDIUM;
913 if (bdrv_check_request(bs, sector_num, nb_sectors))
914 return -EIO;
915
916 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
917 }
918
919 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
920 int nb_sectors, int dirty)
921 {
922 int64_t start, end;
923 unsigned long val, idx, bit;
924
925 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
926 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
927
928 for (; start <= end; start++) {
929 idx = start / (sizeof(unsigned long) * 8);
930 bit = start % (sizeof(unsigned long) * 8);
931 val = bs->dirty_bitmap[idx];
932 if (dirty) {
933 if (!(val & (1UL << bit))) {
934 bs->dirty_count++;
935 val |= 1UL << bit;
936 }
937 } else {
938 if (val & (1UL << bit)) {
939 bs->dirty_count--;
940 val &= ~(1UL << bit);
941 }
942 }
943 bs->dirty_bitmap[idx] = val;
944 }
945 }
946
947 /* Return < 0 if error. Important errors are:
948 -EIO generic I/O error (may happen for all errors)
949 -ENOMEDIUM No media inserted.
950 -EINVAL Invalid sector number or nb_sectors
951 -EACCES Trying to write a read-only device
952 */
953 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
954 const uint8_t *buf, int nb_sectors)
955 {
956 BlockDriver *drv = bs->drv;
957 if (!bs->drv)
958 return -ENOMEDIUM;
959 if (bs->read_only)
960 return -EACCES;
961 if (bdrv_check_request(bs, sector_num, nb_sectors))
962 return -EIO;
963
964 if (bs->dirty_bitmap) {
965 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
966 }
967
968 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
969 bs->wr_highest_sector = sector_num + nb_sectors - 1;
970 }
971
972 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
973 }
974
975 int bdrv_pread(BlockDriverState *bs, int64_t offset,
976 void *buf, int count1)
977 {
978 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
979 int len, nb_sectors, count;
980 int64_t sector_num;
981 int ret;
982
983 count = count1;
984 /* first read to align to sector start */
985 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
986 if (len > count)
987 len = count;
988 sector_num = offset >> BDRV_SECTOR_BITS;
989 if (len > 0) {
990 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
991 return ret;
992 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
993 count -= len;
994 if (count == 0)
995 return count1;
996 sector_num++;
997 buf += len;
998 }
999
1000 /* read the sectors "in place" */
1001 nb_sectors = count >> BDRV_SECTOR_BITS;
1002 if (nb_sectors > 0) {
1003 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1004 return ret;
1005 sector_num += nb_sectors;
1006 len = nb_sectors << BDRV_SECTOR_BITS;
1007 buf += len;
1008 count -= len;
1009 }
1010
1011 /* add data from the last sector */
1012 if (count > 0) {
1013 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1014 return ret;
1015 memcpy(buf, tmp_buf, count);
1016 }
1017 return count1;
1018 }
1019
1020 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1021 const void *buf, int count1)
1022 {
1023 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1024 int len, nb_sectors, count;
1025 int64_t sector_num;
1026 int ret;
1027
1028 count = count1;
1029 /* first write to align to sector start */
1030 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1031 if (len > count)
1032 len = count;
1033 sector_num = offset >> BDRV_SECTOR_BITS;
1034 if (len > 0) {
1035 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1036 return ret;
1037 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1038 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1039 return ret;
1040 count -= len;
1041 if (count == 0)
1042 return count1;
1043 sector_num++;
1044 buf += len;
1045 }
1046
1047 /* write the sectors "in place" */
1048 nb_sectors = count >> BDRV_SECTOR_BITS;
1049 if (nb_sectors > 0) {
1050 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1051 return ret;
1052 sector_num += nb_sectors;
1053 len = nb_sectors << BDRV_SECTOR_BITS;
1054 buf += len;
1055 count -= len;
1056 }
1057
1058 /* add data from the last sector */
1059 if (count > 0) {
1060 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1061 return ret;
1062 memcpy(tmp_buf, buf, count);
1063 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1064 return ret;
1065 }
1066 return count1;
1067 }
1068
1069 /*
1070 * Writes to the file and ensures that no writes are reordered across this
1071 * request (acts as a barrier)
1072 *
1073 * Returns 0 on success, -errno in error cases.
1074 */
1075 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1076 const void *buf, int count)
1077 {
1078 int ret;
1079
1080 ret = bdrv_pwrite(bs, offset, buf, count);
1081 if (ret < 0) {
1082 return ret;
1083 }
1084
1085 /* No flush needed for cache=writethrough, it uses O_DSYNC */
1086 if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1087 bdrv_flush(bs);
1088 }
1089
1090 return 0;
1091 }
1092
1093 /*
1094 * Writes to the file and ensures that no writes are reordered across this
1095 * request (acts as a barrier)
1096 *
1097 * Returns 0 on success, -errno in error cases.
1098 */
1099 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1100 const uint8_t *buf, int nb_sectors)
1101 {
1102 return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1103 buf, BDRV_SECTOR_SIZE * nb_sectors);
1104 }
1105
1106 /**
1107 * Truncate file to 'offset' bytes (needed only for file protocols)
1108 */
1109 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1110 {
1111 BlockDriver *drv = bs->drv;
1112 int ret;
1113 if (!drv)
1114 return -ENOMEDIUM;
1115 if (!drv->bdrv_truncate)
1116 return -ENOTSUP;
1117 if (bs->read_only)
1118 return -EACCES;
1119 ret = drv->bdrv_truncate(bs, offset);
1120 if (ret == 0) {
1121 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1122 }
1123 return ret;
1124 }
1125
1126 /**
1127 * Length of a file in bytes. Return < 0 if error or unknown.
1128 */
1129 int64_t bdrv_getlength(BlockDriverState *bs)
1130 {
1131 BlockDriver *drv = bs->drv;
1132 if (!drv)
1133 return -ENOMEDIUM;
1134
1135 /* Fixed size devices use the total_sectors value for speed instead of
1136 issuing a length query (like lseek) on each call. Also, legacy block
1137 drivers don't provide a bdrv_getlength function and must use
1138 total_sectors. */
1139 if (!bs->growable || !drv->bdrv_getlength) {
1140 return bs->total_sectors * BDRV_SECTOR_SIZE;
1141 }
1142 return drv->bdrv_getlength(bs);
1143 }
1144
1145 /* return 0 as number of sectors if no device present or error */
1146 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1147 {
1148 int64_t length;
1149 length = bdrv_getlength(bs);
1150 if (length < 0)
1151 length = 0;
1152 else
1153 length = length >> BDRV_SECTOR_BITS;
1154 *nb_sectors_ptr = length;
1155 }
1156
1157 struct partition {
1158 uint8_t boot_ind; /* 0x80 - active */
1159 uint8_t head; /* starting head */
1160 uint8_t sector; /* starting sector */
1161 uint8_t cyl; /* starting cylinder */
1162 uint8_t sys_ind; /* What partition type */
1163 uint8_t end_head; /* end head */
1164 uint8_t end_sector; /* end sector */
1165 uint8_t end_cyl; /* end cylinder */
1166 uint32_t start_sect; /* starting sector counting from 0 */
1167 uint32_t nr_sects; /* nr of sectors in partition */
1168 } __attribute__((packed));
1169
1170 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1171 static int guess_disk_lchs(BlockDriverState *bs,
1172 int *pcylinders, int *pheads, int *psectors)
1173 {
1174 uint8_t buf[BDRV_SECTOR_SIZE];
1175 int ret, i, heads, sectors, cylinders;
1176 struct partition *p;
1177 uint32_t nr_sects;
1178 uint64_t nb_sectors;
1179
1180 bdrv_get_geometry(bs, &nb_sectors);
1181
1182 ret = bdrv_read(bs, 0, buf, 1);
1183 if (ret < 0)
1184 return -1;
1185 /* test msdos magic */
1186 if (buf[510] != 0x55 || buf[511] != 0xaa)
1187 return -1;
1188 for(i = 0; i < 4; i++) {
1189 p = ((struct partition *)(buf + 0x1be)) + i;
1190 nr_sects = le32_to_cpu(p->nr_sects);
1191 if (nr_sects && p->end_head) {
1192 /* We make the assumption that the partition terminates on
1193 a cylinder boundary */
1194 heads = p->end_head + 1;
1195 sectors = p->end_sector & 63;
1196 if (sectors == 0)
1197 continue;
1198 cylinders = nb_sectors / (heads * sectors);
1199 if (cylinders < 1 || cylinders > 16383)
1200 continue;
1201 *pheads = heads;
1202 *psectors = sectors;
1203 *pcylinders = cylinders;
1204 #if 0
1205 printf("guessed geometry: LCHS=%d %d %d\n",
1206 cylinders, heads, sectors);
1207 #endif
1208 return 0;
1209 }
1210 }
1211 return -1;
1212 }
1213
1214 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1215 {
1216 int translation, lba_detected = 0;
1217 int cylinders, heads, secs;
1218 uint64_t nb_sectors;
1219
1220 /* if a geometry hint is available, use it */
1221 bdrv_get_geometry(bs, &nb_sectors);
1222 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1223 translation = bdrv_get_translation_hint(bs);
1224 if (cylinders != 0) {
1225 *pcyls = cylinders;
1226 *pheads = heads;
1227 *psecs = secs;
1228 } else {
1229 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1230 if (heads > 16) {
1231 /* if heads > 16, it means that a BIOS LBA
1232 translation was active, so the default
1233 hardware geometry is OK */
1234 lba_detected = 1;
1235 goto default_geometry;
1236 } else {
1237 *pcyls = cylinders;
1238 *pheads = heads;
1239 *psecs = secs;
1240 /* disable any translation to be in sync with
1241 the logical geometry */
1242 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1243 bdrv_set_translation_hint(bs,
1244 BIOS_ATA_TRANSLATION_NONE);
1245 }
1246 }
1247 } else {
1248 default_geometry:
1249 /* if no geometry, use a standard physical disk geometry */
1250 cylinders = nb_sectors / (16 * 63);
1251
1252 if (cylinders > 16383)
1253 cylinders = 16383;
1254 else if (cylinders < 2)
1255 cylinders = 2;
1256 *pcyls = cylinders;
1257 *pheads = 16;
1258 *psecs = 63;
1259 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1260 if ((*pcyls * *pheads) <= 131072) {
1261 bdrv_set_translation_hint(bs,
1262 BIOS_ATA_TRANSLATION_LARGE);
1263 } else {
1264 bdrv_set_translation_hint(bs,
1265 BIOS_ATA_TRANSLATION_LBA);
1266 }
1267 }
1268 }
1269 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1270 }
1271 }
1272
1273 void bdrv_set_geometry_hint(BlockDriverState *bs,
1274 int cyls, int heads, int secs)
1275 {
1276 bs->cyls = cyls;
1277 bs->heads = heads;
1278 bs->secs = secs;
1279 }
1280
1281 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1282 {
1283 bs->type = type;
1284 bs->removable = ((type == BDRV_TYPE_CDROM ||
1285 type == BDRV_TYPE_FLOPPY));
1286 }
1287
1288 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1289 {
1290 bs->translation = translation;
1291 }
1292
1293 void bdrv_get_geometry_hint(BlockDriverState *bs,
1294 int *pcyls, int *pheads, int *psecs)
1295 {
1296 *pcyls = bs->cyls;
1297 *pheads = bs->heads;
1298 *psecs = bs->secs;
1299 }
1300
1301 int bdrv_get_type_hint(BlockDriverState *bs)
1302 {
1303 return bs->type;
1304 }
1305
1306 int bdrv_get_translation_hint(BlockDriverState *bs)
1307 {
1308 return bs->translation;
1309 }
1310
1311 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1312 BlockErrorAction on_write_error)
1313 {
1314 bs->on_read_error = on_read_error;
1315 bs->on_write_error = on_write_error;
1316 }
1317
1318 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1319 {
1320 return is_read ? bs->on_read_error : bs->on_write_error;
1321 }
1322
1323 void bdrv_set_removable(BlockDriverState *bs, int removable)
1324 {
1325 bs->removable = removable;
1326 if (removable && bs == bs_snapshots) {
1327 bs_snapshots = NULL;
1328 }
1329 }
1330
1331 int bdrv_is_removable(BlockDriverState *bs)
1332 {
1333 return bs->removable;
1334 }
1335
1336 int bdrv_is_read_only(BlockDriverState *bs)
1337 {
1338 return bs->read_only;
1339 }
1340
1341 int bdrv_is_sg(BlockDriverState *bs)
1342 {
1343 return bs->sg;
1344 }
1345
1346 int bdrv_enable_write_cache(BlockDriverState *bs)
1347 {
1348 return bs->enable_write_cache;
1349 }
1350
1351 /* XXX: no longer used */
1352 void bdrv_set_change_cb(BlockDriverState *bs,
1353 void (*change_cb)(void *opaque), void *opaque)
1354 {
1355 bs->change_cb = change_cb;
1356 bs->change_opaque = opaque;
1357 }
1358
1359 int bdrv_is_encrypted(BlockDriverState *bs)
1360 {
1361 if (bs->backing_hd && bs->backing_hd->encrypted)
1362 return 1;
1363 return bs->encrypted;
1364 }
1365
1366 int bdrv_key_required(BlockDriverState *bs)
1367 {
1368 BlockDriverState *backing_hd = bs->backing_hd;
1369
1370 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1371 return 1;
1372 return (bs->encrypted && !bs->valid_key);
1373 }
1374
1375 int bdrv_set_key(BlockDriverState *bs, const char *key)
1376 {
1377 int ret;
1378 if (bs->backing_hd && bs->backing_hd->encrypted) {
1379 ret = bdrv_set_key(bs->backing_hd, key);
1380 if (ret < 0)
1381 return ret;
1382 if (!bs->encrypted)
1383 return 0;
1384 }
1385 if (!bs->encrypted) {
1386 return -EINVAL;
1387 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1388 return -ENOMEDIUM;
1389 }
1390 ret = bs->drv->bdrv_set_key(bs, key);
1391 if (ret < 0) {
1392 bs->valid_key = 0;
1393 } else if (!bs->valid_key) {
1394 bs->valid_key = 1;
1395 /* call the change callback now, we skipped it on open */
1396 bs->media_changed = 1;
1397 if (bs->change_cb)
1398 bs->change_cb(bs->change_opaque);
1399 }
1400 return ret;
1401 }
1402
1403 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1404 {
1405 if (!bs->drv) {
1406 buf[0] = '\0';
1407 } else {
1408 pstrcpy(buf, buf_size, bs->drv->format_name);
1409 }
1410 }
1411
1412 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1413 void *opaque)
1414 {
1415 BlockDriver *drv;
1416
1417 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1418 it(opaque, drv->format_name);
1419 }
1420 }
1421
1422 BlockDriverState *bdrv_find(const char *name)
1423 {
1424 BlockDriverState *bs;
1425
1426 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1427 if (!strcmp(name, bs->device_name)) {
1428 return bs;
1429 }
1430 }
1431 return NULL;
1432 }
1433
1434 BlockDriverState *bdrv_next(BlockDriverState *bs)
1435 {
1436 if (!bs) {
1437 return QTAILQ_FIRST(&bdrv_states);
1438 }
1439 return QTAILQ_NEXT(bs, list);
1440 }
1441
1442 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1443 {
1444 BlockDriverState *bs;
1445
1446 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1447 it(opaque, bs);
1448 }
1449 }
1450
1451 const char *bdrv_get_device_name(BlockDriverState *bs)
1452 {
1453 return bs->device_name;
1454 }
1455
1456 int bdrv_flush(BlockDriverState *bs)
1457 {
1458 if (bs->open_flags & BDRV_O_NO_FLUSH) {
1459 return 0;
1460 }
1461
1462 if (bs->drv && bs->drv->bdrv_flush) {
1463 return bs->drv->bdrv_flush(bs);
1464 }
1465
1466 /*
1467 * Some block drivers always operate in either writethrough or unsafe mode
1468 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1469 * the server works (because the behaviour is hardcoded or depends on
1470 * server-side configuration), so we can't ensure that everything is safe
1471 * on disk. Returning an error doesn't work because that would break guests
1472 * even if the server operates in writethrough mode.
1473 *
1474 * Let's hope the user knows what he's doing.
1475 */
1476 return 0;
1477 }
1478
1479 void bdrv_flush_all(void)
1480 {
1481 BlockDriverState *bs;
1482
1483 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1484 if (bs->drv && !bdrv_is_read_only(bs) &&
1485 (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1486 bdrv_flush(bs);
1487 }
1488 }
1489 }
1490
1491 int bdrv_has_zero_init(BlockDriverState *bs)
1492 {
1493 assert(bs->drv);
1494
1495 if (bs->drv->bdrv_has_zero_init) {
1496 return bs->drv->bdrv_has_zero_init(bs);
1497 }
1498
1499 return 1;
1500 }
1501
1502 /*
1503 * Returns true iff the specified sector is present in the disk image. Drivers
1504 * not implementing the functionality are assumed to not support backing files,
1505 * hence all their sectors are reported as allocated.
1506 *
1507 * 'pnum' is set to the number of sectors (including and immediately following
1508 * the specified sector) that are known to be in the same
1509 * allocated/unallocated state.
1510 *
1511 * 'nb_sectors' is the max value 'pnum' should be set to.
1512 */
1513 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1514 int *pnum)
1515 {
1516 int64_t n;
1517 if (!bs->drv->bdrv_is_allocated) {
1518 if (sector_num >= bs->total_sectors) {
1519 *pnum = 0;
1520 return 0;
1521 }
1522 n = bs->total_sectors - sector_num;
1523 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1524 return 1;
1525 }
1526 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1527 }
1528
1529 void bdrv_mon_event(const BlockDriverState *bdrv,
1530 BlockMonEventAction action, int is_read)
1531 {
1532 QObject *data;
1533 const char *action_str;
1534
1535 switch (action) {
1536 case BDRV_ACTION_REPORT:
1537 action_str = "report";
1538 break;
1539 case BDRV_ACTION_IGNORE:
1540 action_str = "ignore";
1541 break;
1542 case BDRV_ACTION_STOP:
1543 action_str = "stop";
1544 break;
1545 default:
1546 abort();
1547 }
1548
1549 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1550 bdrv->device_name,
1551 action_str,
1552 is_read ? "read" : "write");
1553 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1554
1555 qobject_decref(data);
1556 }
1557
1558 static void bdrv_print_dict(QObject *obj, void *opaque)
1559 {
1560 QDict *bs_dict;
1561 Monitor *mon = opaque;
1562
1563 bs_dict = qobject_to_qdict(obj);
1564
1565 monitor_printf(mon, "%s: type=%s removable=%d",
1566 qdict_get_str(bs_dict, "device"),
1567 qdict_get_str(bs_dict, "type"),
1568 qdict_get_bool(bs_dict, "removable"));
1569
1570 if (qdict_get_bool(bs_dict, "removable")) {
1571 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1572 }
1573
1574 if (qdict_haskey(bs_dict, "inserted")) {
1575 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1576
1577 monitor_printf(mon, " file=");
1578 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1579 if (qdict_haskey(qdict, "backing_file")) {
1580 monitor_printf(mon, " backing_file=");
1581 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1582 }
1583 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1584 qdict_get_bool(qdict, "ro"),
1585 qdict_get_str(qdict, "drv"),
1586 qdict_get_bool(qdict, "encrypted"));
1587 } else {
1588 monitor_printf(mon, " [not inserted]");
1589 }
1590
1591 monitor_printf(mon, "\n");
1592 }
1593
1594 void bdrv_info_print(Monitor *mon, const QObject *data)
1595 {
1596 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1597 }
1598
1599 void bdrv_info(Monitor *mon, QObject **ret_data)
1600 {
1601 QList *bs_list;
1602 BlockDriverState *bs;
1603
1604 bs_list = qlist_new();
1605
1606 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1607 QObject *bs_obj;
1608 const char *type = "unknown";
1609
1610 switch(bs->type) {
1611 case BDRV_TYPE_HD:
1612 type = "hd";
1613 break;
1614 case BDRV_TYPE_CDROM:
1615 type = "cdrom";
1616 break;
1617 case BDRV_TYPE_FLOPPY:
1618 type = "floppy";
1619 break;
1620 }
1621
1622 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1623 "'removable': %i, 'locked': %i }",
1624 bs->device_name, type, bs->removable,
1625 bs->locked);
1626
1627 if (bs->drv) {
1628 QObject *obj;
1629 QDict *bs_dict = qobject_to_qdict(bs_obj);
1630
1631 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1632 "'encrypted': %i }",
1633 bs->filename, bs->read_only,
1634 bs->drv->format_name,
1635 bdrv_is_encrypted(bs));
1636 if (bs->backing_file[0] != '\0') {
1637 QDict *qdict = qobject_to_qdict(obj);
1638 qdict_put(qdict, "backing_file",
1639 qstring_from_str(bs->backing_file));
1640 }
1641
1642 qdict_put_obj(bs_dict, "inserted", obj);
1643 }
1644 qlist_append_obj(bs_list, bs_obj);
1645 }
1646
1647 *ret_data = QOBJECT(bs_list);
1648 }
1649
1650 static void bdrv_stats_iter(QObject *data, void *opaque)
1651 {
1652 QDict *qdict;
1653 Monitor *mon = opaque;
1654
1655 qdict = qobject_to_qdict(data);
1656 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1657
1658 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1659 monitor_printf(mon, " rd_bytes=%" PRId64
1660 " wr_bytes=%" PRId64
1661 " rd_operations=%" PRId64
1662 " wr_operations=%" PRId64
1663 "\n",
1664 qdict_get_int(qdict, "rd_bytes"),
1665 qdict_get_int(qdict, "wr_bytes"),
1666 qdict_get_int(qdict, "rd_operations"),
1667 qdict_get_int(qdict, "wr_operations"));
1668 }
1669
1670 void bdrv_stats_print(Monitor *mon, const QObject *data)
1671 {
1672 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1673 }
1674
1675 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1676 {
1677 QObject *res;
1678 QDict *dict;
1679
1680 res = qobject_from_jsonf("{ 'stats': {"
1681 "'rd_bytes': %" PRId64 ","
1682 "'wr_bytes': %" PRId64 ","
1683 "'rd_operations': %" PRId64 ","
1684 "'wr_operations': %" PRId64 ","
1685 "'wr_highest_offset': %" PRId64
1686 "} }",
1687 bs->rd_bytes, bs->wr_bytes,
1688 bs->rd_ops, bs->wr_ops,
1689 bs->wr_highest_sector *
1690 (uint64_t)BDRV_SECTOR_SIZE);
1691 dict = qobject_to_qdict(res);
1692
1693 if (*bs->device_name) {
1694 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1695 }
1696
1697 if (bs->file) {
1698 QObject *parent = bdrv_info_stats_bs(bs->file);
1699 qdict_put_obj(dict, "parent", parent);
1700 }
1701
1702 return res;
1703 }
1704
1705 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1706 {
1707 QObject *obj;
1708 QList *devices;
1709 BlockDriverState *bs;
1710
1711 devices = qlist_new();
1712
1713 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1714 obj = bdrv_info_stats_bs(bs);
1715 qlist_append_obj(devices, obj);
1716 }
1717
1718 *ret_data = QOBJECT(devices);
1719 }
1720
1721 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1722 {
1723 if (bs->backing_hd && bs->backing_hd->encrypted)
1724 return bs->backing_file;
1725 else if (bs->encrypted)
1726 return bs->filename;
1727 else
1728 return NULL;
1729 }
1730
1731 void bdrv_get_backing_filename(BlockDriverState *bs,
1732 char *filename, int filename_size)
1733 {
1734 if (!bs->backing_file) {
1735 pstrcpy(filename, filename_size, "");
1736 } else {
1737 pstrcpy(filename, filename_size, bs->backing_file);
1738 }
1739 }
1740
1741 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1742 const uint8_t *buf, int nb_sectors)
1743 {
1744 BlockDriver *drv = bs->drv;
1745 if (!drv)
1746 return -ENOMEDIUM;
1747 if (!drv->bdrv_write_compressed)
1748 return -ENOTSUP;
1749 if (bdrv_check_request(bs, sector_num, nb_sectors))
1750 return -EIO;
1751
1752 if (bs->dirty_bitmap) {
1753 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1754 }
1755
1756 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1757 }
1758
1759 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1760 {
1761 BlockDriver *drv = bs->drv;
1762 if (!drv)
1763 return -ENOMEDIUM;
1764 if (!drv->bdrv_get_info)
1765 return -ENOTSUP;
1766 memset(bdi, 0, sizeof(*bdi));
1767 return drv->bdrv_get_info(bs, bdi);
1768 }
1769
1770 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1771 int64_t pos, int size)
1772 {
1773 BlockDriver *drv = bs->drv;
1774 if (!drv)
1775 return -ENOMEDIUM;
1776 if (drv->bdrv_save_vmstate)
1777 return drv->bdrv_save_vmstate(bs, buf, pos, size);
1778 if (bs->file)
1779 return bdrv_save_vmstate(bs->file, buf, pos, size);
1780 return -ENOTSUP;
1781 }
1782
1783 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1784 int64_t pos, int size)
1785 {
1786 BlockDriver *drv = bs->drv;
1787 if (!drv)
1788 return -ENOMEDIUM;
1789 if (drv->bdrv_load_vmstate)
1790 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1791 if (bs->file)
1792 return bdrv_load_vmstate(bs->file, buf, pos, size);
1793 return -ENOTSUP;
1794 }
1795
1796 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1797 {
1798 BlockDriver *drv = bs->drv;
1799
1800 if (!drv || !drv->bdrv_debug_event) {
1801 return;
1802 }
1803
1804 return drv->bdrv_debug_event(bs, event);
1805
1806 }
1807
1808 /**************************************************************/
1809 /* handling of snapshots */
1810
1811 int bdrv_can_snapshot(BlockDriverState *bs)
1812 {
1813 BlockDriver *drv = bs->drv;
1814 if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1815 return 0;
1816 }
1817
1818 if (!drv->bdrv_snapshot_create) {
1819 if (bs->file != NULL) {
1820 return bdrv_can_snapshot(bs->file);
1821 }
1822 return 0;
1823 }
1824
1825 return 1;
1826 }
1827
1828 int bdrv_is_snapshot(BlockDriverState *bs)
1829 {
1830 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
1831 }
1832
1833 BlockDriverState *bdrv_snapshots(void)
1834 {
1835 BlockDriverState *bs;
1836
1837 if (bs_snapshots) {
1838 return bs_snapshots;
1839 }
1840
1841 bs = NULL;
1842 while ((bs = bdrv_next(bs))) {
1843 if (bdrv_can_snapshot(bs)) {
1844 bs_snapshots = bs;
1845 return bs;
1846 }
1847 }
1848 return NULL;
1849 }
1850
1851 int bdrv_snapshot_create(BlockDriverState *bs,
1852 QEMUSnapshotInfo *sn_info)
1853 {
1854 BlockDriver *drv = bs->drv;
1855 if (!drv)
1856 return -ENOMEDIUM;
1857 if (drv->bdrv_snapshot_create)
1858 return drv->bdrv_snapshot_create(bs, sn_info);
1859 if (bs->file)
1860 return bdrv_snapshot_create(bs->file, sn_info);
1861 return -ENOTSUP;
1862 }
1863
1864 int bdrv_snapshot_goto(BlockDriverState *bs,
1865 const char *snapshot_id)
1866 {
1867 BlockDriver *drv = bs->drv;
1868 int ret, open_ret;
1869
1870 if (!drv)
1871 return -ENOMEDIUM;
1872 if (drv->bdrv_snapshot_goto)
1873 return drv->bdrv_snapshot_goto(bs, snapshot_id);
1874
1875 if (bs->file) {
1876 drv->bdrv_close(bs);
1877 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1878 open_ret = drv->bdrv_open(bs, bs->open_flags);
1879 if (open_ret < 0) {
1880 bdrv_delete(bs->file);
1881 bs->drv = NULL;
1882 return open_ret;
1883 }
1884 return ret;
1885 }
1886
1887 return -ENOTSUP;
1888 }
1889
1890 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1891 {
1892 BlockDriver *drv = bs->drv;
1893 if (!drv)
1894 return -ENOMEDIUM;
1895 if (drv->bdrv_snapshot_delete)
1896 return drv->bdrv_snapshot_delete(bs, snapshot_id);
1897 if (bs->file)
1898 return bdrv_snapshot_delete(bs->file, snapshot_id);
1899 return -ENOTSUP;
1900 }
1901
1902 int bdrv_snapshot_list(BlockDriverState *bs,
1903 QEMUSnapshotInfo **psn_info)
1904 {
1905 BlockDriver *drv = bs->drv;
1906 if (!drv)
1907 return -ENOMEDIUM;
1908 if (drv->bdrv_snapshot_list)
1909 return drv->bdrv_snapshot_list(bs, psn_info);
1910 if (bs->file)
1911 return bdrv_snapshot_list(bs->file, psn_info);
1912 return -ENOTSUP;
1913 }
1914
1915 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
1916 const char *snapshot_name)
1917 {
1918 BlockDriver *drv = bs->drv;
1919 if (!drv) {
1920 return -ENOMEDIUM;
1921 }
1922 if (!bs->read_only) {
1923 return -EINVAL;
1924 }
1925 if (drv->bdrv_snapshot_load_tmp) {
1926 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
1927 }
1928 return -ENOTSUP;
1929 }
1930
1931 #define NB_SUFFIXES 4
1932
1933 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1934 {
1935 static const char suffixes[NB_SUFFIXES] = "KMGT";
1936 int64_t base;
1937 int i;
1938
1939 if (size <= 999) {
1940 snprintf(buf, buf_size, "%" PRId64, size);
1941 } else {
1942 base = 1024;
1943 for(i = 0; i < NB_SUFFIXES; i++) {
1944 if (size < (10 * base)) {
1945 snprintf(buf, buf_size, "%0.1f%c",
1946 (double)size / base,
1947 suffixes[i]);
1948 break;
1949 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1950 snprintf(buf, buf_size, "%" PRId64 "%c",
1951 ((size + (base >> 1)) / base),
1952 suffixes[i]);
1953 break;
1954 }
1955 base = base * 1024;
1956 }
1957 }
1958 return buf;
1959 }
1960
1961 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1962 {
1963 char buf1[128], date_buf[128], clock_buf[128];
1964 #ifdef _WIN32
1965 struct tm *ptm;
1966 #else
1967 struct tm tm;
1968 #endif
1969 time_t ti;
1970 int64_t secs;
1971
1972 if (!sn) {
1973 snprintf(buf, buf_size,
1974 "%-10s%-20s%7s%20s%15s",
1975 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1976 } else {
1977 ti = sn->date_sec;
1978 #ifdef _WIN32
1979 ptm = localtime(&ti);
1980 strftime(date_buf, sizeof(date_buf),
1981 "%Y-%m-%d %H:%M:%S", ptm);
1982 #else
1983 localtime_r(&ti, &tm);
1984 strftime(date_buf, sizeof(date_buf),
1985 "%Y-%m-%d %H:%M:%S", &tm);
1986 #endif
1987 secs = sn->vm_clock_nsec / 1000000000;
1988 snprintf(clock_buf, sizeof(clock_buf),
1989 "%02d:%02d:%02d.%03d",
1990 (int)(secs / 3600),
1991 (int)((secs / 60) % 60),
1992 (int)(secs % 60),
1993 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1994 snprintf(buf, buf_size,
1995 "%-10s%-20s%7s%20s%15s",
1996 sn->id_str, sn->name,
1997 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1998 date_buf,
1999 clock_buf);
2000 }
2001 return buf;
2002 }
2003
2004
2005 /**************************************************************/
2006 /* async I/Os */
2007
2008 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2009 QEMUIOVector *qiov, int nb_sectors,
2010 BlockDriverCompletionFunc *cb, void *opaque)
2011 {
2012 BlockDriver *drv = bs->drv;
2013 BlockDriverAIOCB *ret;
2014
2015 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2016
2017 if (!drv)
2018 return NULL;
2019 if (bdrv_check_request(bs, sector_num, nb_sectors))
2020 return NULL;
2021
2022 ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2023 cb, opaque);
2024
2025 if (ret) {
2026 /* Update stats even though technically transfer has not happened. */
2027 bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2028 bs->rd_ops ++;
2029 }
2030
2031 return ret;
2032 }
2033
2034 typedef struct BlockCompleteData {
2035 BlockDriverCompletionFunc *cb;
2036 void *opaque;
2037 BlockDriverState *bs;
2038 int64_t sector_num;
2039 int nb_sectors;
2040 } BlockCompleteData;
2041
2042 static void block_complete_cb(void *opaque, int ret)
2043 {
2044 BlockCompleteData *b = opaque;
2045
2046 if (b->bs->dirty_bitmap) {
2047 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2048 }
2049 b->cb(b->opaque, ret);
2050 qemu_free(b);
2051 }
2052
2053 static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2054 int64_t sector_num,
2055 int nb_sectors,
2056 BlockDriverCompletionFunc *cb,
2057 void *opaque)
2058 {
2059 BlockCompleteData *blkdata = qemu_mallocz(sizeof(BlockCompleteData));
2060
2061 blkdata->bs = bs;
2062 blkdata->cb = cb;
2063 blkdata->opaque = opaque;
2064 blkdata->sector_num = sector_num;
2065 blkdata->nb_sectors = nb_sectors;
2066
2067 return blkdata;
2068 }
2069
2070 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2071 QEMUIOVector *qiov, int nb_sectors,
2072 BlockDriverCompletionFunc *cb, void *opaque)
2073 {
2074 BlockDriver *drv = bs->drv;
2075 BlockDriverAIOCB *ret;
2076 BlockCompleteData *blk_cb_data;
2077
2078 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2079
2080 if (!drv)
2081 return NULL;
2082 if (bs->read_only)
2083 return NULL;
2084 if (bdrv_check_request(bs, sector_num, nb_sectors))
2085 return NULL;
2086
2087 if (bs->dirty_bitmap) {
2088 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2089 opaque);
2090 cb = &block_complete_cb;
2091 opaque = blk_cb_data;
2092 }
2093
2094 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2095 cb, opaque);
2096
2097 if (ret) {
2098 /* Update stats even though technically transfer has not happened. */
2099 bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2100 bs->wr_ops ++;
2101 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2102 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2103 }
2104 }
2105
2106 return ret;
2107 }
2108
2109
2110 typedef struct MultiwriteCB {
2111 int error;
2112 int num_requests;
2113 int num_callbacks;
2114 struct {
2115 BlockDriverCompletionFunc *cb;
2116 void *opaque;
2117 QEMUIOVector *free_qiov;
2118 void *free_buf;
2119 } callbacks[];
2120 } MultiwriteCB;
2121
2122 static void multiwrite_user_cb(MultiwriteCB *mcb)
2123 {
2124 int i;
2125
2126 for (i = 0; i < mcb->num_callbacks; i++) {
2127 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2128 if (mcb->callbacks[i].free_qiov) {
2129 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2130 }
2131 qemu_free(mcb->callbacks[i].free_qiov);
2132 qemu_vfree(mcb->callbacks[i].free_buf);
2133 }
2134 }
2135
2136 static void multiwrite_cb(void *opaque, int ret)
2137 {
2138 MultiwriteCB *mcb = opaque;
2139
2140 trace_multiwrite_cb(mcb, ret);
2141
2142 if (ret < 0 && !mcb->error) {
2143 mcb->error = ret;
2144 }
2145
2146 mcb->num_requests--;
2147 if (mcb->num_requests == 0) {
2148 multiwrite_user_cb(mcb);
2149 qemu_free(mcb);
2150 }
2151 }
2152
2153 static int multiwrite_req_compare(const void *a, const void *b)
2154 {
2155 const BlockRequest *req1 = a, *req2 = b;
2156
2157 /*
2158 * Note that we can't simply subtract req2->sector from req1->sector
2159 * here as that could overflow the return value.
2160 */
2161 if (req1->sector > req2->sector) {
2162 return 1;
2163 } else if (req1->sector < req2->sector) {
2164 return -1;
2165 } else {
2166 return 0;
2167 }
2168 }
2169
2170 /*
2171 * Takes a bunch of requests and tries to merge them. Returns the number of
2172 * requests that remain after merging.
2173 */
2174 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2175 int num_reqs, MultiwriteCB *mcb)
2176 {
2177 int i, outidx;
2178
2179 // Sort requests by start sector
2180 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2181
2182 // Check if adjacent requests touch the same clusters. If so, combine them,
2183 // filling up gaps with zero sectors.
2184 outidx = 0;
2185 for (i = 1; i < num_reqs; i++) {
2186 int merge = 0;
2187 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2188
2189 // This handles the cases that are valid for all block drivers, namely
2190 // exactly sequential writes and overlapping writes.
2191 if (reqs[i].sector <= oldreq_last) {
2192 merge = 1;
2193 }
2194
2195 // The block driver may decide that it makes sense to combine requests
2196 // even if there is a gap of some sectors between them. In this case,
2197 // the gap is filled with zeros (therefore only applicable for yet
2198 // unused space in format like qcow2).
2199 if (!merge && bs->drv->bdrv_merge_requests) {
2200 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2201 }
2202
2203 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2204 merge = 0;
2205 }
2206
2207 if (merge) {
2208 size_t size;
2209 QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2210 qemu_iovec_init(qiov,
2211 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2212
2213 // Add the first request to the merged one. If the requests are
2214 // overlapping, drop the last sectors of the first request.
2215 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2216 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2217
2218 // We might need to add some zeros between the two requests
2219 if (reqs[i].sector > oldreq_last) {
2220 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2221 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2222 memset(buf, 0, zero_bytes);
2223 qemu_iovec_add(qiov, buf, zero_bytes);
2224 mcb->callbacks[i].free_buf = buf;
2225 }
2226
2227 // Add the second request
2228 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2229
2230 reqs[outidx].nb_sectors = qiov->size >> 9;
2231 reqs[outidx].qiov = qiov;
2232
2233 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2234 } else {
2235 outidx++;
2236 reqs[outidx].sector = reqs[i].sector;
2237 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2238 reqs[outidx].qiov = reqs[i].qiov;
2239 }
2240 }
2241
2242 return outidx + 1;
2243 }
2244
2245 /*
2246 * Submit multiple AIO write requests at once.
2247 *
2248 * On success, the function returns 0 and all requests in the reqs array have
2249 * been submitted. In error case this function returns -1, and any of the
2250 * requests may or may not be submitted yet. In particular, this means that the
2251 * callback will be called for some of the requests, for others it won't. The
2252 * caller must check the error field of the BlockRequest to wait for the right
2253 * callbacks (if error != 0, no callback will be called).
2254 *
2255 * The implementation may modify the contents of the reqs array, e.g. to merge
2256 * requests. However, the fields opaque and error are left unmodified as they
2257 * are used to signal failure for a single request to the caller.
2258 */
2259 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2260 {
2261 BlockDriverAIOCB *acb;
2262 MultiwriteCB *mcb;
2263 int i;
2264
2265 if (num_reqs == 0) {
2266 return 0;
2267 }
2268
2269 // Create MultiwriteCB structure
2270 mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2271 mcb->num_requests = 0;
2272 mcb->num_callbacks = num_reqs;
2273
2274 for (i = 0; i < num_reqs; i++) {
2275 mcb->callbacks[i].cb = reqs[i].cb;
2276 mcb->callbacks[i].opaque = reqs[i].opaque;
2277 }
2278
2279 // Check for mergable requests
2280 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2281
2282 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2283
2284 /*
2285 * Run the aio requests. As soon as one request can't be submitted
2286 * successfully, fail all requests that are not yet submitted (we must
2287 * return failure for all requests anyway)
2288 *
2289 * num_requests cannot be set to the right value immediately: If
2290 * bdrv_aio_writev fails for some request, num_requests would be too high
2291 * and therefore multiwrite_cb() would never recognize the multiwrite
2292 * request as completed. We also cannot use the loop variable i to set it
2293 * when the first request fails because the callback may already have been
2294 * called for previously submitted requests. Thus, num_requests must be
2295 * incremented for each request that is submitted.
2296 *
2297 * The problem that callbacks may be called early also means that we need
2298 * to take care that num_requests doesn't become 0 before all requests are
2299 * submitted - multiwrite_cb() would consider the multiwrite request
2300 * completed. A dummy request that is "completed" by a manual call to
2301 * multiwrite_cb() takes care of this.
2302 */
2303 mcb->num_requests = 1;
2304
2305 // Run the aio requests
2306 for (i = 0; i < num_reqs; i++) {
2307 mcb->num_requests++;
2308 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2309 reqs[i].nb_sectors, multiwrite_cb, mcb);
2310
2311 if (acb == NULL) {
2312 // We can only fail the whole thing if no request has been
2313 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2314 // complete and report the error in the callback.
2315 if (i == 0) {
2316 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2317 goto fail;
2318 } else {
2319 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2320 multiwrite_cb(mcb, -EIO);
2321 break;
2322 }
2323 }
2324 }
2325
2326 /* Complete the dummy request */
2327 multiwrite_cb(mcb, 0);
2328
2329 return 0;
2330
2331 fail:
2332 for (i = 0; i < mcb->num_callbacks; i++) {
2333 reqs[i].error = -EIO;
2334 }
2335 qemu_free(mcb);
2336 return -1;
2337 }
2338
2339 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2340 BlockDriverCompletionFunc *cb, void *opaque)
2341 {
2342 BlockDriver *drv = bs->drv;
2343
2344 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2345 return bdrv_aio_noop_em(bs, cb, opaque);
2346 }
2347
2348 if (!drv)
2349 return NULL;
2350 return drv->bdrv_aio_flush(bs, cb, opaque);
2351 }
2352
2353 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2354 {
2355 acb->pool->cancel(acb);
2356 }
2357
2358
2359 /**************************************************************/
2360 /* async block device emulation */
2361
2362 typedef struct BlockDriverAIOCBSync {
2363 BlockDriverAIOCB common;
2364 QEMUBH *bh;
2365 int ret;
2366 /* vector translation state */
2367 QEMUIOVector *qiov;
2368 uint8_t *bounce;
2369 int is_write;
2370 } BlockDriverAIOCBSync;
2371
2372 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2373 {
2374 BlockDriverAIOCBSync *acb =
2375 container_of(blockacb, BlockDriverAIOCBSync, common);
2376 qemu_bh_delete(acb->bh);
2377 acb->bh = NULL;
2378 qemu_aio_release(acb);
2379 }
2380
2381 static AIOPool bdrv_em_aio_pool = {
2382 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2383 .cancel = bdrv_aio_cancel_em,
2384 };
2385
2386 static void bdrv_aio_bh_cb(void *opaque)
2387 {
2388 BlockDriverAIOCBSync *acb = opaque;
2389
2390 if (!acb->is_write)
2391 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2392 qemu_vfree(acb->bounce);
2393 acb->common.cb(acb->common.opaque, acb->ret);
2394 qemu_bh_delete(acb->bh);
2395 acb->bh = NULL;
2396 qemu_aio_release(acb);
2397 }
2398
2399 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2400 int64_t sector_num,
2401 QEMUIOVector *qiov,
2402 int nb_sectors,
2403 BlockDriverCompletionFunc *cb,
2404 void *opaque,
2405 int is_write)
2406
2407 {
2408 BlockDriverAIOCBSync *acb;
2409
2410 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2411 acb->is_write = is_write;
2412 acb->qiov = qiov;
2413 acb->bounce = qemu_blockalign(bs, qiov->size);
2414
2415 if (!acb->bh)
2416 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2417
2418 if (is_write) {
2419 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2420 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2421 } else {
2422 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2423 }
2424
2425 qemu_bh_schedule(acb->bh);
2426
2427 return &acb->common;
2428 }
2429
2430 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2431 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2432 BlockDriverCompletionFunc *cb, void *opaque)
2433 {
2434 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2435 }
2436
2437 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2438 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2439 BlockDriverCompletionFunc *cb, void *opaque)
2440 {
2441 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2442 }
2443
2444 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2445 BlockDriverCompletionFunc *cb, void *opaque)
2446 {
2447 BlockDriverAIOCBSync *acb;
2448
2449 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2450 acb->is_write = 1; /* don't bounce in the completion hadler */
2451 acb->qiov = NULL;
2452 acb->bounce = NULL;
2453 acb->ret = 0;
2454
2455 if (!acb->bh)
2456 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2457
2458 bdrv_flush(bs);
2459 qemu_bh_schedule(acb->bh);
2460 return &acb->common;
2461 }
2462
2463 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2464 BlockDriverCompletionFunc *cb, void *opaque)
2465 {
2466 BlockDriverAIOCBSync *acb;
2467
2468 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2469 acb->is_write = 1; /* don't bounce in the completion handler */
2470 acb->qiov = NULL;
2471 acb->bounce = NULL;
2472 acb->ret = 0;
2473
2474 if (!acb->bh) {
2475 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2476 }
2477
2478 qemu_bh_schedule(acb->bh);
2479 return &acb->common;
2480 }
2481
2482 /**************************************************************/
2483 /* sync block device emulation */
2484
2485 static void bdrv_rw_em_cb(void *opaque, int ret)
2486 {
2487 *(int *)opaque = ret;
2488 }
2489
2490 #define NOT_DONE 0x7fffffff
2491
2492 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2493 uint8_t *buf, int nb_sectors)
2494 {
2495 int async_ret;
2496 BlockDriverAIOCB *acb;
2497 struct iovec iov;
2498 QEMUIOVector qiov;
2499
2500 async_context_push();
2501
2502 async_ret = NOT_DONE;
2503 iov.iov_base = (void *)buf;
2504 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2505 qemu_iovec_init_external(&qiov, &iov, 1);
2506 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2507 bdrv_rw_em_cb, &async_ret);
2508 if (acb == NULL) {
2509 async_ret = -1;
2510 goto fail;
2511 }
2512
2513 while (async_ret == NOT_DONE) {
2514 qemu_aio_wait();
2515 }
2516
2517
2518 fail:
2519 async_context_pop();
2520 return async_ret;
2521 }
2522
2523 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2524 const uint8_t *buf, int nb_sectors)
2525 {
2526 int async_ret;
2527 BlockDriverAIOCB *acb;
2528 struct iovec iov;
2529 QEMUIOVector qiov;
2530
2531 async_context_push();
2532
2533 async_ret = NOT_DONE;
2534 iov.iov_base = (void *)buf;
2535 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2536 qemu_iovec_init_external(&qiov, &iov, 1);
2537 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2538 bdrv_rw_em_cb, &async_ret);
2539 if (acb == NULL) {
2540 async_ret = -1;
2541 goto fail;
2542 }
2543 while (async_ret == NOT_DONE) {
2544 qemu_aio_wait();
2545 }
2546
2547 fail:
2548 async_context_pop();
2549 return async_ret;
2550 }
2551
2552 void bdrv_init(void)
2553 {
2554 module_call_init(MODULE_INIT_BLOCK);
2555 }
2556
2557 void bdrv_init_with_whitelist(void)
2558 {
2559 use_bdrv_whitelist = 1;
2560 bdrv_init();
2561 }
2562
2563 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2564 BlockDriverCompletionFunc *cb, void *opaque)
2565 {
2566 BlockDriverAIOCB *acb;
2567
2568 if (pool->free_aiocb) {
2569 acb = pool->free_aiocb;
2570 pool->free_aiocb = acb->next;
2571 } else {
2572 acb = qemu_mallocz(pool->aiocb_size);
2573 acb->pool = pool;
2574 }
2575 acb->bs = bs;
2576 acb->cb = cb;
2577 acb->opaque = opaque;
2578 return acb;
2579 }
2580
2581 void qemu_aio_release(void *p)
2582 {
2583 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2584 AIOPool *pool = acb->pool;
2585 acb->next = pool->free_aiocb;
2586 pool->free_aiocb = acb;
2587 }
2588
2589 /**************************************************************/
2590 /* removable device support */
2591
2592 /**
2593 * Return TRUE if the media is present
2594 */
2595 int bdrv_is_inserted(BlockDriverState *bs)
2596 {
2597 BlockDriver *drv = bs->drv;
2598 int ret;
2599 if (!drv)
2600 return 0;
2601 if (!drv->bdrv_is_inserted)
2602 return !bs->tray_open;
2603 ret = drv->bdrv_is_inserted(bs);
2604 return ret;
2605 }
2606
2607 /**
2608 * Return TRUE if the media changed since the last call to this
2609 * function. It is currently only used for floppy disks
2610 */
2611 int bdrv_media_changed(BlockDriverState *bs)
2612 {
2613 BlockDriver *drv = bs->drv;
2614 int ret;
2615
2616 if (!drv || !drv->bdrv_media_changed)
2617 ret = -ENOTSUP;
2618 else
2619 ret = drv->bdrv_media_changed(bs);
2620 if (ret == -ENOTSUP)
2621 ret = bs->media_changed;
2622 bs->media_changed = 0;
2623 return ret;
2624 }
2625
2626 /**
2627 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2628 */
2629 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2630 {
2631 BlockDriver *drv = bs->drv;
2632 int ret;
2633
2634 if (bs->locked) {
2635 return -EBUSY;
2636 }
2637
2638 if (!drv || !drv->bdrv_eject) {
2639 ret = -ENOTSUP;
2640 } else {
2641 ret = drv->bdrv_eject(bs, eject_flag);
2642 }
2643 if (ret == -ENOTSUP) {
2644 ret = 0;
2645 }
2646 if (ret >= 0) {
2647 bs->tray_open = eject_flag;
2648 }
2649
2650 return ret;
2651 }
2652
2653 int bdrv_is_locked(BlockDriverState *bs)
2654 {
2655 return bs->locked;
2656 }
2657
2658 /**
2659 * Lock or unlock the media (if it is locked, the user won't be able
2660 * to eject it manually).
2661 */
2662 void bdrv_set_locked(BlockDriverState *bs, int locked)
2663 {
2664 BlockDriver *drv = bs->drv;
2665
2666 bs->locked = locked;
2667 if (drv && drv->bdrv_set_locked) {
2668 drv->bdrv_set_locked(bs, locked);
2669 }
2670 }
2671
2672 /* needed for generic scsi interface */
2673
2674 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2675 {
2676 BlockDriver *drv = bs->drv;
2677
2678 if (drv && drv->bdrv_ioctl)
2679 return drv->bdrv_ioctl(bs, req, buf);
2680 return -ENOTSUP;
2681 }
2682
2683 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2684 unsigned long int req, void *buf,
2685 BlockDriverCompletionFunc *cb, void *opaque)
2686 {
2687 BlockDriver *drv = bs->drv;
2688
2689 if (drv && drv->bdrv_aio_ioctl)
2690 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2691 return NULL;
2692 }
2693
2694
2695
2696 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2697 {
2698 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2699 }
2700
2701 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2702 {
2703 int64_t bitmap_size;
2704
2705 bs->dirty_count = 0;
2706 if (enable) {
2707 if (!bs->dirty_bitmap) {
2708 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2709 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2710 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2711
2712 bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2713 }
2714 } else {
2715 if (bs->dirty_bitmap) {
2716 qemu_free(bs->dirty_bitmap);
2717 bs->dirty_bitmap = NULL;
2718 }
2719 }
2720 }
2721
2722 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2723 {
2724 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2725
2726 if (bs->dirty_bitmap &&
2727 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2728 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2729 (1UL << (chunk % (sizeof(unsigned long) * 8))));
2730 } else {
2731 return 0;
2732 }
2733 }
2734
2735 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2736 int nr_sectors)
2737 {
2738 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2739 }
2740
2741 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2742 {
2743 return bs->dirty_count;
2744 }
QEmu