?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'spice/spice.v45' into staging
[qemu.git] / block.c
blob70aab63e4b8edcde7661c31c92d0b811bf2bb70e
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 #include "qemu-coroutine.h"
32
33 #ifdef CONFIG_BSD
34 #include <sys/types.h>
35 #include <sys/stat.h>
36 #include <sys/ioctl.h>
37 #include <sys/queue.h>
38 #ifndef __DragonFly__
39 #include <sys/disk.h>
40 #endif
41 #endif
42
43 #ifdef _WIN32
44 #include <windows.h>
45 #endif
46
47 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
48
49 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
50 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
51 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
52 BlockDriverCompletionFunc *cb, void *opaque);
53 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
54 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
55 BlockDriverCompletionFunc *cb, void *opaque);
56 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
57 int64_t sector_num, int nb_sectors,
58 QEMUIOVector *iov);
59 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
60 int64_t sector_num, int nb_sectors,
61 QEMUIOVector *iov);
62 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
63 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
64 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
65 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
66 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
67 int64_t sector_num,
68 QEMUIOVector *qiov,
69 int nb_sectors,
70 BlockDriverCompletionFunc *cb,
71 void *opaque,
72 bool is_write);
73 static void coroutine_fn bdrv_co_do_rw(void *opaque);
74
75 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
76 QTAILQ_HEAD_INITIALIZER(bdrv_states);
77
78 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
79 QLIST_HEAD_INITIALIZER(bdrv_drivers);
80
81 /* The device to use for VM snapshots */
82 static BlockDriverState *bs_snapshots;
83
84 /* If non-zero, use only whitelisted block drivers */
85 static int use_bdrv_whitelist;
86
87 #ifdef _WIN32
88 static int is_windows_drive_prefix(const char *filename)
89 {
90 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
91 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
92 filename[1] == ':');
93 }
94
95 int is_windows_drive(const char *filename)
96 {
97 if (is_windows_drive_prefix(filename) &&
98 filename[2] == '\0')
99 return 1;
100 if (strstart(filename, "\\\\.\\", NULL) ||
101 strstart(filename, "//./", NULL))
102 return 1;
103 return 0;
104 }
105 #endif
106
107 /* check if the path starts with "<protocol>:" */
108 static int path_has_protocol(const char *path)
109 {
110 #ifdef _WIN32
111 if (is_windows_drive(path) ||
112 is_windows_drive_prefix(path)) {
113 return 0;
114 }
115 #endif
116
117 return strchr(path, ':') != NULL;
118 }
119
120 int path_is_absolute(const char *path)
121 {
122 const char *p;
123 #ifdef _WIN32
124 /* specific case for names like: "\\.\d:" */
125 if (*path == '/' || *path == '\\')
126 return 1;
127 #endif
128 p = strchr(path, ':');
129 if (p)
130 p++;
131 else
132 p = path;
133 #ifdef _WIN32
134 return (*p == '/' || *p == '\\');
135 #else
136 return (*p == '/');
137 #endif
138 }
139
140 /* if filename is absolute, just copy it to dest. Otherwise, build a
141 path to it by considering it is relative to base_path. URL are
142 supported. */
143 void path_combine(char *dest, int dest_size,
144 const char *base_path,
145 const char *filename)
146 {
147 const char *p, *p1;
148 int len;
149
150 if (dest_size <= 0)
151 return;
152 if (path_is_absolute(filename)) {
153 pstrcpy(dest, dest_size, filename);
154 } else {
155 p = strchr(base_path, ':');
156 if (p)
157 p++;
158 else
159 p = base_path;
160 p1 = strrchr(base_path, '/');
161 #ifdef _WIN32
162 {
163 const char *p2;
164 p2 = strrchr(base_path, '\\');
165 if (!p1 || p2 > p1)
166 p1 = p2;
167 }
168 #endif
169 if (p1)
170 p1++;
171 else
172 p1 = base_path;
173 if (p1 > p)
174 p = p1;
175 len = p - base_path;
176 if (len > dest_size - 1)
177 len = dest_size - 1;
178 memcpy(dest, base_path, len);
179 dest[len] = '\0';
180 pstrcat(dest, dest_size, filename);
181 }
182 }
183
184 void bdrv_register(BlockDriver *bdrv)
185 {
186 /* Block drivers without coroutine functions need emulation */
187 if (!bdrv->bdrv_co_readv) {
188 bdrv->bdrv_co_readv = bdrv_co_readv_em;
189 bdrv->bdrv_co_writev = bdrv_co_writev_em;
190
191 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
192 * the block driver lacks aio we need to emulate that too.
193 */
194 if (!bdrv->bdrv_aio_readv) {
195 /* add AIO emulation layer */
196 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
197 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
198 }
199 }
200
201 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
202 }
203
204 /* create a new block device (by default it is empty) */
205 BlockDriverState *bdrv_new(const char *device_name)
206 {
207 BlockDriverState *bs;
208
209 bs = g_malloc0(sizeof(BlockDriverState));
210 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
211 if (device_name[0] != '\0') {
212 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
213 }
214 bdrv_iostatus_disable(bs);
215 return bs;
216 }
217
218 BlockDriver *bdrv_find_format(const char *format_name)
219 {
220 BlockDriver *drv1;
221 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
222 if (!strcmp(drv1->format_name, format_name)) {
223 return drv1;
224 }
225 }
226 return NULL;
227 }
228
229 static int bdrv_is_whitelisted(BlockDriver *drv)
230 {
231 static const char *whitelist[] = {
232 CONFIG_BDRV_WHITELIST
233 };
234 const char **p;
235
236 if (!whitelist[0])
237 return 1; /* no whitelist, anything goes */
238
239 for (p = whitelist; *p; p++) {
240 if (!strcmp(drv->format_name, *p)) {
241 return 1;
242 }
243 }
244 return 0;
245 }
246
247 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
248 {
249 BlockDriver *drv = bdrv_find_format(format_name);
250 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
251 }
252
253 int bdrv_create(BlockDriver *drv, const char* filename,
254 QEMUOptionParameter *options)
255 {
256 if (!drv->bdrv_create)
257 return -ENOTSUP;
258
259 return drv->bdrv_create(filename, options);
260 }
261
262 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
263 {
264 BlockDriver *drv;
265
266 drv = bdrv_find_protocol(filename);
267 if (drv == NULL) {
268 return -ENOENT;
269 }
270
271 return bdrv_create(drv, filename, options);
272 }
273
274 #ifdef _WIN32
275 void get_tmp_filename(char *filename, int size)
276 {
277 char temp_dir[MAX_PATH];
278
279 GetTempPath(MAX_PATH, temp_dir);
280 GetTempFileName(temp_dir, "qem", 0, filename);
281 }
282 #else
283 void get_tmp_filename(char *filename, int size)
284 {
285 int fd;
286 const char *tmpdir;
287 /* XXX: race condition possible */
288 tmpdir = getenv("TMPDIR");
289 if (!tmpdir)
290 tmpdir = "/tmp";
291 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
292 fd = mkstemp(filename);
293 close(fd);
294 }
295 #endif
296
297 /*
298 * Detect host devices. By convention, /dev/cdrom[N] is always
299 * recognized as a host CDROM.
300 */
301 static BlockDriver *find_hdev_driver(const char *filename)
302 {
303 int score_max = 0, score;
304 BlockDriver *drv = NULL, *d;
305
306 QLIST_FOREACH(d, &bdrv_drivers, list) {
307 if (d->bdrv_probe_device) {
308 score = d->bdrv_probe_device(filename);
309 if (score > score_max) {
310 score_max = score;
311 drv = d;
312 }
313 }
314 }
315
316 return drv;
317 }
318
319 BlockDriver *bdrv_find_protocol(const char *filename)
320 {
321 BlockDriver *drv1;
322 char protocol[128];
323 int len;
324 const char *p;
325
326 /* TODO Drivers without bdrv_file_open must be specified explicitly */
327
328 /*
329 * XXX(hch): we really should not let host device detection
330 * override an explicit protocol specification, but moving this
331 * later breaks access to device names with colons in them.
332 * Thanks to the brain-dead persistent naming schemes on udev-
333 * based Linux systems those actually are quite common.
334 */
335 drv1 = find_hdev_driver(filename);
336 if (drv1) {
337 return drv1;
338 }
339
340 if (!path_has_protocol(filename)) {
341 return bdrv_find_format("file");
342 }
343 p = strchr(filename, ':');
344 assert(p != NULL);
345 len = p - filename;
346 if (len > sizeof(protocol) - 1)
347 len = sizeof(protocol) - 1;
348 memcpy(protocol, filename, len);
349 protocol[len] = '\0';
350 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
351 if (drv1->protocol_name &&
352 !strcmp(drv1->protocol_name, protocol)) {
353 return drv1;
354 }
355 }
356 return NULL;
357 }
358
359 static int find_image_format(const char *filename, BlockDriver **pdrv)
360 {
361 int ret, score, score_max;
362 BlockDriver *drv1, *drv;
363 uint8_t buf[2048];
364 BlockDriverState *bs;
365
366 ret = bdrv_file_open(&bs, filename, 0);
367 if (ret < 0) {
368 *pdrv = NULL;
369 return ret;
370 }
371
372 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
373 if (bs->sg || !bdrv_is_inserted(bs)) {
374 bdrv_delete(bs);
375 drv = bdrv_find_format("raw");
376 if (!drv) {
377 ret = -ENOENT;
378 }
379 *pdrv = drv;
380 return ret;
381 }
382
383 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
384 bdrv_delete(bs);
385 if (ret < 0) {
386 *pdrv = NULL;
387 return ret;
388 }
389
390 score_max = 0;
391 drv = NULL;
392 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
393 if (drv1->bdrv_probe) {
394 score = drv1->bdrv_probe(buf, ret, filename);
395 if (score > score_max) {
396 score_max = score;
397 drv = drv1;
398 }
399 }
400 }
401 if (!drv) {
402 ret = -ENOENT;
403 }
404 *pdrv = drv;
405 return ret;
406 }
407
408 /**
409 * Set the current 'total_sectors' value
410 */
411 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
412 {
413 BlockDriver *drv = bs->drv;
414
415 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
416 if (bs->sg)
417 return 0;
418
419 /* query actual device if possible, otherwise just trust the hint */
420 if (drv->bdrv_getlength) {
421 int64_t length = drv->bdrv_getlength(bs);
422 if (length < 0) {
423 return length;
424 }
425 hint = length >> BDRV_SECTOR_BITS;
426 }
427
428 bs->total_sectors = hint;
429 return 0;
430 }
431
432 /**
433 * Set open flags for a given cache mode
434 *
435 * Return 0 on success, -1 if the cache mode was invalid.
436 */
437 int bdrv_parse_cache_flags(const char *mode, int *flags)
438 {
439 *flags &= ~BDRV_O_CACHE_MASK;
440
441 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
442 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
443 } else if (!strcmp(mode, "directsync")) {
444 *flags |= BDRV_O_NOCACHE;
445 } else if (!strcmp(mode, "writeback")) {
446 *flags |= BDRV_O_CACHE_WB;
447 } else if (!strcmp(mode, "unsafe")) {
448 *flags |= BDRV_O_CACHE_WB;
449 *flags |= BDRV_O_NO_FLUSH;
450 } else if (!strcmp(mode, "writethrough")) {
451 /* this is the default */
452 } else {
453 return -1;
454 }
455
456 return 0;
457 }
458
459 /*
460 * Common part for opening disk images and files
461 */
462 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
463 int flags, BlockDriver *drv)
464 {
465 int ret, open_flags;
466
467 assert(drv != NULL);
468
469 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
470
471 bs->file = NULL;
472 bs->total_sectors = 0;
473 bs->encrypted = 0;
474 bs->valid_key = 0;
475 bs->open_flags = flags;
476 bs->buffer_alignment = 512;
477
478 pstrcpy(bs->filename, sizeof(bs->filename), filename);
479
480 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
481 return -ENOTSUP;
482 }
483
484 bs->drv = drv;
485 bs->opaque = g_malloc0(drv->instance_size);
486
487 if (flags & BDRV_O_CACHE_WB)
488 bs->enable_write_cache = 1;
489
490 /*
491 * Clear flags that are internal to the block layer before opening the
492 * image.
493 */
494 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
495
496 /*
497 * Snapshots should be writable.
498 */
499 if (bs->is_temporary) {
500 open_flags |= BDRV_O_RDWR;
501 }
502
503 /* Open the image, either directly or using a protocol */
504 if (drv->bdrv_file_open) {
505 ret = drv->bdrv_file_open(bs, filename, open_flags);
506 } else {
507 ret = bdrv_file_open(&bs->file, filename, open_flags);
508 if (ret >= 0) {
509 ret = drv->bdrv_open(bs, open_flags);
510 }
511 }
512
513 if (ret < 0) {
514 goto free_and_fail;
515 }
516
517 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
518
519 ret = refresh_total_sectors(bs, bs->total_sectors);
520 if (ret < 0) {
521 goto free_and_fail;
522 }
523
524 #ifndef _WIN32
525 if (bs->is_temporary) {
526 unlink(filename);
527 }
528 #endif
529 return 0;
530
531 free_and_fail:
532 if (bs->file) {
533 bdrv_delete(bs->file);
534 bs->file = NULL;
535 }
536 g_free(bs->opaque);
537 bs->opaque = NULL;
538 bs->drv = NULL;
539 return ret;
540 }
541
542 /*
543 * Opens a file using a protocol (file, host_device, nbd, ...)
544 */
545 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
546 {
547 BlockDriverState *bs;
548 BlockDriver *drv;
549 int ret;
550
551 drv = bdrv_find_protocol(filename);
552 if (!drv) {
553 return -ENOENT;
554 }
555
556 bs = bdrv_new("");
557 ret = bdrv_open_common(bs, filename, flags, drv);
558 if (ret < 0) {
559 bdrv_delete(bs);
560 return ret;
561 }
562 bs->growable = 1;
563 *pbs = bs;
564 return 0;
565 }
566
567 /*
568 * Opens a disk image (raw, qcow2, vmdk, ...)
569 */
570 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
571 BlockDriver *drv)
572 {
573 int ret;
574
575 if (flags & BDRV_O_SNAPSHOT) {
576 BlockDriverState *bs1;
577 int64_t total_size;
578 int is_protocol = 0;
579 BlockDriver *bdrv_qcow2;
580 QEMUOptionParameter *options;
581 char tmp_filename[PATH_MAX];
582 char backing_filename[PATH_MAX];
583
584 /* if snapshot, we create a temporary backing file and open it
585 instead of opening 'filename' directly */
586
587 /* if there is a backing file, use it */
588 bs1 = bdrv_new("");
589 ret = bdrv_open(bs1, filename, 0, drv);
590 if (ret < 0) {
591 bdrv_delete(bs1);
592 return ret;
593 }
594 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
595
596 if (bs1->drv && bs1->drv->protocol_name)
597 is_protocol = 1;
598
599 bdrv_delete(bs1);
600
601 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
602
603 /* Real path is meaningless for protocols */
604 if (is_protocol)
605 snprintf(backing_filename, sizeof(backing_filename),
606 "%s", filename);
607 else if (!realpath(filename, backing_filename))
608 return -errno;
609
610 bdrv_qcow2 = bdrv_find_format("qcow2");
611 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
612
613 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
614 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
615 if (drv) {
616 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
617 drv->format_name);
618 }
619
620 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
621 free_option_parameters(options);
622 if (ret < 0) {
623 return ret;
624 }
625
626 filename = tmp_filename;
627 drv = bdrv_qcow2;
628 bs->is_temporary = 1;
629 }
630
631 /* Find the right image format driver */
632 if (!drv) {
633 ret = find_image_format(filename, &drv);
634 }
635
636 if (!drv) {
637 goto unlink_and_fail;
638 }
639
640 /* Open the image */
641 ret = bdrv_open_common(bs, filename, flags, drv);
642 if (ret < 0) {
643 goto unlink_and_fail;
644 }
645
646 /* If there is a backing file, use it */
647 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
648 char backing_filename[PATH_MAX];
649 int back_flags;
650 BlockDriver *back_drv = NULL;
651
652 bs->backing_hd = bdrv_new("");
653
654 if (path_has_protocol(bs->backing_file)) {
655 pstrcpy(backing_filename, sizeof(backing_filename),
656 bs->backing_file);
657 } else {
658 path_combine(backing_filename, sizeof(backing_filename),
659 filename, bs->backing_file);
660 }
661
662 if (bs->backing_format[0] != '\0') {
663 back_drv = bdrv_find_format(bs->backing_format);
664 }
665
666 /* backing files always opened read-only */
667 back_flags =
668 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
669
670 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
671 if (ret < 0) {
672 bdrv_close(bs);
673 return ret;
674 }
675 if (bs->is_temporary) {
676 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
677 } else {
678 /* base image inherits from "parent" */
679 bs->backing_hd->keep_read_only = bs->keep_read_only;
680 }
681 }
682
683 if (!bdrv_key_required(bs)) {
684 bdrv_dev_change_media_cb(bs, true);
685 }
686
687 return 0;
688
689 unlink_and_fail:
690 if (bs->is_temporary) {
691 unlink(filename);
692 }
693 return ret;
694 }
695
696 void bdrv_close(BlockDriverState *bs)
697 {
698 if (bs->drv) {
699 if (bs == bs_snapshots) {
700 bs_snapshots = NULL;
701 }
702 if (bs->backing_hd) {
703 bdrv_delete(bs->backing_hd);
704 bs->backing_hd = NULL;
705 }
706 bs->drv->bdrv_close(bs);
707 g_free(bs->opaque);
708 #ifdef _WIN32
709 if (bs->is_temporary) {
710 unlink(bs->filename);
711 }
712 #endif
713 bs->opaque = NULL;
714 bs->drv = NULL;
715
716 if (bs->file != NULL) {
717 bdrv_close(bs->file);
718 }
719
720 bdrv_dev_change_media_cb(bs, false);
721 }
722 }
723
724 void bdrv_close_all(void)
725 {
726 BlockDriverState *bs;
727
728 QTAILQ_FOREACH(bs, &bdrv_states, list) {
729 bdrv_close(bs);
730 }
731 }
732
733 /* make a BlockDriverState anonymous by removing from bdrv_state list.
734 Also, NULL terminate the device_name to prevent double remove */
735 void bdrv_make_anon(BlockDriverState *bs)
736 {
737 if (bs->device_name[0] != '\0') {
738 QTAILQ_REMOVE(&bdrv_states, bs, list);
739 }
740 bs->device_name[0] = '\0';
741 }
742
743 void bdrv_delete(BlockDriverState *bs)
744 {
745 assert(!bs->dev);
746
747 /* remove from list, if necessary */
748 bdrv_make_anon(bs);
749
750 bdrv_close(bs);
751 if (bs->file != NULL) {
752 bdrv_delete(bs->file);
753 }
754
755 assert(bs != bs_snapshots);
756 g_free(bs);
757 }
758
759 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
760 /* TODO change to DeviceState *dev when all users are qdevified */
761 {
762 if (bs->dev) {
763 return -EBUSY;
764 }
765 bs->dev = dev;
766 bdrv_iostatus_reset(bs);
767 return 0;
768 }
769
770 /* TODO qdevified devices don't use this, remove when devices are qdevified */
771 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
772 {
773 if (bdrv_attach_dev(bs, dev) < 0) {
774 abort();
775 }
776 }
777
778 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
779 /* TODO change to DeviceState *dev when all users are qdevified */
780 {
781 assert(bs->dev == dev);
782 bs->dev = NULL;
783 bs->dev_ops = NULL;
784 bs->dev_opaque = NULL;
785 bs->buffer_alignment = 512;
786 }
787
788 /* TODO change to return DeviceState * when all users are qdevified */
789 void *bdrv_get_attached_dev(BlockDriverState *bs)
790 {
791 return bs->dev;
792 }
793
794 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
795 void *opaque)
796 {
797 bs->dev_ops = ops;
798 bs->dev_opaque = opaque;
799 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
800 bs_snapshots = NULL;
801 }
802 }
803
804 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
805 {
806 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
807 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
808 }
809 }
810
811 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
812 {
813 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
814 }
815
816 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
817 {
818 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
819 return bs->dev_ops->is_tray_open(bs->dev_opaque);
820 }
821 return false;
822 }
823
824 static void bdrv_dev_resize_cb(BlockDriverState *bs)
825 {
826 if (bs->dev_ops && bs->dev_ops->resize_cb) {
827 bs->dev_ops->resize_cb(bs->dev_opaque);
828 }
829 }
830
831 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
832 {
833 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
834 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
835 }
836 return false;
837 }
838
839 /*
840 * Run consistency checks on an image
841 *
842 * Returns 0 if the check could be completed (it doesn't mean that the image is
843 * free of errors) or -errno when an internal error occurred. The results of the
844 * check are stored in res.
845 */
846 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
847 {
848 if (bs->drv->bdrv_check == NULL) {
849 return -ENOTSUP;
850 }
851
852 memset(res, 0, sizeof(*res));
853 return bs->drv->bdrv_check(bs, res);
854 }
855
856 #define COMMIT_BUF_SECTORS 2048
857
858 /* commit COW file into the raw image */
859 int bdrv_commit(BlockDriverState *bs)
860 {
861 BlockDriver *drv = bs->drv;
862 BlockDriver *backing_drv;
863 int64_t sector, total_sectors;
864 int n, ro, open_flags;
865 int ret = 0, rw_ret = 0;
866 uint8_t *buf;
867 char filename[1024];
868 BlockDriverState *bs_rw, *bs_ro;
869
870 if (!drv)
871 return -ENOMEDIUM;
872
873 if (!bs->backing_hd) {
874 return -ENOTSUP;
875 }
876
877 if (bs->backing_hd->keep_read_only) {
878 return -EACCES;
879 }
880
881 backing_drv = bs->backing_hd->drv;
882 ro = bs->backing_hd->read_only;
883 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
884 open_flags = bs->backing_hd->open_flags;
885
886 if (ro) {
887 /* re-open as RW */
888 bdrv_delete(bs->backing_hd);
889 bs->backing_hd = NULL;
890 bs_rw = bdrv_new("");
891 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
892 backing_drv);
893 if (rw_ret < 0) {
894 bdrv_delete(bs_rw);
895 /* try to re-open read-only */
896 bs_ro = bdrv_new("");
897 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
898 backing_drv);
899 if (ret < 0) {
900 bdrv_delete(bs_ro);
901 /* drive not functional anymore */
902 bs->drv = NULL;
903 return ret;
904 }
905 bs->backing_hd = bs_ro;
906 return rw_ret;
907 }
908 bs->backing_hd = bs_rw;
909 }
910
911 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
912 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
913
914 for (sector = 0; sector < total_sectors; sector += n) {
915 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
916
917 if (bdrv_read(bs, sector, buf, n) != 0) {
918 ret = -EIO;
919 goto ro_cleanup;
920 }
921
922 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
923 ret = -EIO;
924 goto ro_cleanup;
925 }
926 }
927 }
928
929 if (drv->bdrv_make_empty) {
930 ret = drv->bdrv_make_empty(bs);
931 bdrv_flush(bs);
932 }
933
934 /*
935 * Make sure all data we wrote to the backing device is actually
936 * stable on disk.
937 */
938 if (bs->backing_hd)
939 bdrv_flush(bs->backing_hd);
940
941 ro_cleanup:
942 g_free(buf);
943
944 if (ro) {
945 /* re-open as RO */
946 bdrv_delete(bs->backing_hd);
947 bs->backing_hd = NULL;
948 bs_ro = bdrv_new("");
949 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
950 backing_drv);
951 if (ret < 0) {
952 bdrv_delete(bs_ro);
953 /* drive not functional anymore */
954 bs->drv = NULL;
955 return ret;
956 }
957 bs->backing_hd = bs_ro;
958 bs->backing_hd->keep_read_only = 0;
959 }
960
961 return ret;
962 }
963
964 void bdrv_commit_all(void)
965 {
966 BlockDriverState *bs;
967
968 QTAILQ_FOREACH(bs, &bdrv_states, list) {
969 bdrv_commit(bs);
970 }
971 }
972
973 /*
974 * Return values:
975 * 0 - success
976 * -EINVAL - backing format specified, but no file
977 * -ENOSPC - can't update the backing file because no space is left in the
978 * image file header
979 * -ENOTSUP - format driver doesn't support changing the backing file
980 */
981 int bdrv_change_backing_file(BlockDriverState *bs,
982 const char *backing_file, const char *backing_fmt)
983 {
984 BlockDriver *drv = bs->drv;
985
986 if (drv->bdrv_change_backing_file != NULL) {
987 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
988 } else {
989 return -ENOTSUP;
990 }
991 }
992
993 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
994 size_t size)
995 {
996 int64_t len;
997
998 if (!bdrv_is_inserted(bs))
999 return -ENOMEDIUM;
1000
1001 if (bs->growable)
1002 return 0;
1003
1004 len = bdrv_getlength(bs);
1005
1006 if (offset < 0)
1007 return -EIO;
1008
1009 if ((offset > len) || (len - offset < size))
1010 return -EIO;
1011
1012 return 0;
1013 }
1014
1015 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1016 int nb_sectors)
1017 {
1018 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1019 nb_sectors * BDRV_SECTOR_SIZE);
1020 }
1021
1022 typedef struct RwCo {
1023 BlockDriverState *bs;
1024 int64_t sector_num;
1025 int nb_sectors;
1026 QEMUIOVector *qiov;
1027 bool is_write;
1028 int ret;
1029 } RwCo;
1030
1031 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1032 {
1033 RwCo *rwco = opaque;
1034
1035 if (!rwco->is_write) {
1036 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1037 rwco->nb_sectors, rwco->qiov);
1038 } else {
1039 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1040 rwco->nb_sectors, rwco->qiov);
1041 }
1042 }
1043
1044 /*
1045 * Process a synchronous request using coroutines
1046 */
1047 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1048 int nb_sectors, bool is_write)
1049 {
1050 QEMUIOVector qiov;
1051 struct iovec iov = {
1052 .iov_base = (void *)buf,
1053 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1054 };
1055 Coroutine *co;
1056 RwCo rwco = {
1057 .bs = bs,
1058 .sector_num = sector_num,
1059 .nb_sectors = nb_sectors,
1060 .qiov = &qiov,
1061 .is_write = is_write,
1062 .ret = NOT_DONE,
1063 };
1064
1065 qemu_iovec_init_external(&qiov, &iov, 1);
1066
1067 if (qemu_in_coroutine()) {
1068 /* Fast-path if already in coroutine context */
1069 bdrv_rw_co_entry(&rwco);
1070 } else {
1071 co = qemu_coroutine_create(bdrv_rw_co_entry);
1072 qemu_coroutine_enter(co, &rwco);
1073 while (rwco.ret == NOT_DONE) {
1074 qemu_aio_wait();
1075 }
1076 }
1077 return rwco.ret;
1078 }
1079
1080 /* return < 0 if error. See bdrv_write() for the return codes */
1081 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1082 uint8_t *buf, int nb_sectors)
1083 {
1084 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1085 }
1086
1087 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1088 int nb_sectors, int dirty)
1089 {
1090 int64_t start, end;
1091 unsigned long val, idx, bit;
1092
1093 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1094 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1095
1096 for (; start <= end; start++) {
1097 idx = start / (sizeof(unsigned long) * 8);
1098 bit = start % (sizeof(unsigned long) * 8);
1099 val = bs->dirty_bitmap[idx];
1100 if (dirty) {
1101 if (!(val & (1UL << bit))) {
1102 bs->dirty_count++;
1103 val |= 1UL << bit;
1104 }
1105 } else {
1106 if (val & (1UL << bit)) {
1107 bs->dirty_count--;
1108 val &= ~(1UL << bit);
1109 }
1110 }
1111 bs->dirty_bitmap[idx] = val;
1112 }
1113 }
1114
1115 /* Return < 0 if error. Important errors are:
1116 -EIO generic I/O error (may happen for all errors)
1117 -ENOMEDIUM No media inserted.
1118 -EINVAL Invalid sector number or nb_sectors
1119 -EACCES Trying to write a read-only device
1120 */
1121 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1122 const uint8_t *buf, int nb_sectors)
1123 {
1124 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1125 }
1126
1127 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1128 void *buf, int count1)
1129 {
1130 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1131 int len, nb_sectors, count;
1132 int64_t sector_num;
1133 int ret;
1134
1135 count = count1;
1136 /* first read to align to sector start */
1137 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1138 if (len > count)
1139 len = count;
1140 sector_num = offset >> BDRV_SECTOR_BITS;
1141 if (len > 0) {
1142 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1143 return ret;
1144 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1145 count -= len;
1146 if (count == 0)
1147 return count1;
1148 sector_num++;
1149 buf += len;
1150 }
1151
1152 /* read the sectors "in place" */
1153 nb_sectors = count >> BDRV_SECTOR_BITS;
1154 if (nb_sectors > 0) {
1155 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1156 return ret;
1157 sector_num += nb_sectors;
1158 len = nb_sectors << BDRV_SECTOR_BITS;
1159 buf += len;
1160 count -= len;
1161 }
1162
1163 /* add data from the last sector */
1164 if (count > 0) {
1165 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1166 return ret;
1167 memcpy(buf, tmp_buf, count);
1168 }
1169 return count1;
1170 }
1171
1172 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1173 const void *buf, int count1)
1174 {
1175 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1176 int len, nb_sectors, count;
1177 int64_t sector_num;
1178 int ret;
1179
1180 count = count1;
1181 /* first write to align to sector start */
1182 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1183 if (len > count)
1184 len = count;
1185 sector_num = offset >> BDRV_SECTOR_BITS;
1186 if (len > 0) {
1187 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1188 return ret;
1189 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1190 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1191 return ret;
1192 count -= len;
1193 if (count == 0)
1194 return count1;
1195 sector_num++;
1196 buf += len;
1197 }
1198
1199 /* write the sectors "in place" */
1200 nb_sectors = count >> BDRV_SECTOR_BITS;
1201 if (nb_sectors > 0) {
1202 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1203 return ret;
1204 sector_num += nb_sectors;
1205 len = nb_sectors << BDRV_SECTOR_BITS;
1206 buf += len;
1207 count -= len;
1208 }
1209
1210 /* add data from the last sector */
1211 if (count > 0) {
1212 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1213 return ret;
1214 memcpy(tmp_buf, buf, count);
1215 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1216 return ret;
1217 }
1218 return count1;
1219 }
1220
1221 /*
1222 * Writes to the file and ensures that no writes are reordered across this
1223 * request (acts as a barrier)
1224 *
1225 * Returns 0 on success, -errno in error cases.
1226 */
1227 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1228 const void *buf, int count)
1229 {
1230 int ret;
1231
1232 ret = bdrv_pwrite(bs, offset, buf, count);
1233 if (ret < 0) {
1234 return ret;
1235 }
1236
1237 /* No flush needed for cache modes that use O_DSYNC */
1238 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1239 bdrv_flush(bs);
1240 }
1241
1242 return 0;
1243 }
1244
1245 /*
1246 * Handle a read request in coroutine context
1247 */
1248 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1249 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1250 {
1251 BlockDriver *drv = bs->drv;
1252
1253 if (!drv) {
1254 return -ENOMEDIUM;
1255 }
1256 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1257 return -EIO;
1258 }
1259
1260 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1261 }
1262
1263 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1264 int nb_sectors, QEMUIOVector *qiov)
1265 {
1266 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1267
1268 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov);
1269 }
1270
1271 /*
1272 * Handle a write request in coroutine context
1273 */
1274 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1275 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1276 {
1277 BlockDriver *drv = bs->drv;
1278 int ret;
1279
1280 if (!bs->drv) {
1281 return -ENOMEDIUM;
1282 }
1283 if (bs->read_only) {
1284 return -EACCES;
1285 }
1286 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1287 return -EIO;
1288 }
1289
1290 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1291
1292 if (bs->dirty_bitmap) {
1293 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1294 }
1295
1296 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1297 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1298 }
1299
1300 return ret;
1301 }
1302
1303 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1304 int nb_sectors, QEMUIOVector *qiov)
1305 {
1306 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1307
1308 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov);
1309 }
1310
1311 /**
1312 * Truncate file to 'offset' bytes (needed only for file protocols)
1313 */
1314 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1315 {
1316 BlockDriver *drv = bs->drv;
1317 int ret;
1318 if (!drv)
1319 return -ENOMEDIUM;
1320 if (!drv->bdrv_truncate)
1321 return -ENOTSUP;
1322 if (bs->read_only)
1323 return -EACCES;
1324 if (bdrv_in_use(bs))
1325 return -EBUSY;
1326 ret = drv->bdrv_truncate(bs, offset);
1327 if (ret == 0) {
1328 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1329 bdrv_dev_resize_cb(bs);
1330 }
1331 return ret;
1332 }
1333
1334 /**
1335 * Length of a allocated file in bytes. Sparse files are counted by actual
1336 * allocated space. Return < 0 if error or unknown.
1337 */
1338 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1339 {
1340 BlockDriver *drv = bs->drv;
1341 if (!drv) {
1342 return -ENOMEDIUM;
1343 }
1344 if (drv->bdrv_get_allocated_file_size) {
1345 return drv->bdrv_get_allocated_file_size(bs);
1346 }
1347 if (bs->file) {
1348 return bdrv_get_allocated_file_size(bs->file);
1349 }
1350 return -ENOTSUP;
1351 }
1352
1353 /**
1354 * Length of a file in bytes. Return < 0 if error or unknown.
1355 */
1356 int64_t bdrv_getlength(BlockDriverState *bs)
1357 {
1358 BlockDriver *drv = bs->drv;
1359 if (!drv)
1360 return -ENOMEDIUM;
1361
1362 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
1363 if (drv->bdrv_getlength) {
1364 return drv->bdrv_getlength(bs);
1365 }
1366 }
1367 return bs->total_sectors * BDRV_SECTOR_SIZE;
1368 }
1369
1370 /* return 0 as number of sectors if no device present or error */
1371 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1372 {
1373 int64_t length;
1374 length = bdrv_getlength(bs);
1375 if (length < 0)
1376 length = 0;
1377 else
1378 length = length >> BDRV_SECTOR_BITS;
1379 *nb_sectors_ptr = length;
1380 }
1381
1382 struct partition {
1383 uint8_t boot_ind; /* 0x80 - active */
1384 uint8_t head; /* starting head */
1385 uint8_t sector; /* starting sector */
1386 uint8_t cyl; /* starting cylinder */
1387 uint8_t sys_ind; /* What partition type */
1388 uint8_t end_head; /* end head */
1389 uint8_t end_sector; /* end sector */
1390 uint8_t end_cyl; /* end cylinder */
1391 uint32_t start_sect; /* starting sector counting from 0 */
1392 uint32_t nr_sects; /* nr of sectors in partition */
1393 } QEMU_PACKED;
1394
1395 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1396 static int guess_disk_lchs(BlockDriverState *bs,
1397 int *pcylinders, int *pheads, int *psectors)
1398 {
1399 uint8_t buf[BDRV_SECTOR_SIZE];
1400 int ret, i, heads, sectors, cylinders;
1401 struct partition *p;
1402 uint32_t nr_sects;
1403 uint64_t nb_sectors;
1404
1405 bdrv_get_geometry(bs, &nb_sectors);
1406
1407 ret = bdrv_read(bs, 0, buf, 1);
1408 if (ret < 0)
1409 return -1;
1410 /* test msdos magic */
1411 if (buf[510] != 0x55 || buf[511] != 0xaa)
1412 return -1;
1413 for(i = 0; i < 4; i++) {
1414 p = ((struct partition *)(buf + 0x1be)) + i;
1415 nr_sects = le32_to_cpu(p->nr_sects);
1416 if (nr_sects && p->end_head) {
1417 /* We make the assumption that the partition terminates on
1418 a cylinder boundary */
1419 heads = p->end_head + 1;
1420 sectors = p->end_sector & 63;
1421 if (sectors == 0)
1422 continue;
1423 cylinders = nb_sectors / (heads * sectors);
1424 if (cylinders < 1 || cylinders > 16383)
1425 continue;
1426 *pheads = heads;
1427 *psectors = sectors;
1428 *pcylinders = cylinders;
1429 #if 0
1430 printf("guessed geometry: LCHS=%d %d %d\n",
1431 cylinders, heads, sectors);
1432 #endif
1433 return 0;
1434 }
1435 }
1436 return -1;
1437 }
1438
1439 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1440 {
1441 int translation, lba_detected = 0;
1442 int cylinders, heads, secs;
1443 uint64_t nb_sectors;
1444
1445 /* if a geometry hint is available, use it */
1446 bdrv_get_geometry(bs, &nb_sectors);
1447 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1448 translation = bdrv_get_translation_hint(bs);
1449 if (cylinders != 0) {
1450 *pcyls = cylinders;
1451 *pheads = heads;
1452 *psecs = secs;
1453 } else {
1454 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1455 if (heads > 16) {
1456 /* if heads > 16, it means that a BIOS LBA
1457 translation was active, so the default
1458 hardware geometry is OK */
1459 lba_detected = 1;
1460 goto default_geometry;
1461 } else {
1462 *pcyls = cylinders;
1463 *pheads = heads;
1464 *psecs = secs;
1465 /* disable any translation to be in sync with
1466 the logical geometry */
1467 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1468 bdrv_set_translation_hint(bs,
1469 BIOS_ATA_TRANSLATION_NONE);
1470 }
1471 }
1472 } else {
1473 default_geometry:
1474 /* if no geometry, use a standard physical disk geometry */
1475 cylinders = nb_sectors / (16 * 63);
1476
1477 if (cylinders > 16383)
1478 cylinders = 16383;
1479 else if (cylinders < 2)
1480 cylinders = 2;
1481 *pcyls = cylinders;
1482 *pheads = 16;
1483 *psecs = 63;
1484 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1485 if ((*pcyls * *pheads) <= 131072) {
1486 bdrv_set_translation_hint(bs,
1487 BIOS_ATA_TRANSLATION_LARGE);
1488 } else {
1489 bdrv_set_translation_hint(bs,
1490 BIOS_ATA_TRANSLATION_LBA);
1491 }
1492 }
1493 }
1494 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1495 }
1496 }
1497
1498 void bdrv_set_geometry_hint(BlockDriverState *bs,
1499 int cyls, int heads, int secs)
1500 {
1501 bs->cyls = cyls;
1502 bs->heads = heads;
1503 bs->secs = secs;
1504 }
1505
1506 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1507 {
1508 bs->translation = translation;
1509 }
1510
1511 void bdrv_get_geometry_hint(BlockDriverState *bs,
1512 int *pcyls, int *pheads, int *psecs)
1513 {
1514 *pcyls = bs->cyls;
1515 *pheads = bs->heads;
1516 *psecs = bs->secs;
1517 }
1518
1519 /* Recognize floppy formats */
1520 typedef struct FDFormat {
1521 FDriveType drive;
1522 uint8_t last_sect;
1523 uint8_t max_track;
1524 uint8_t max_head;
1525 } FDFormat;
1526
1527 static const FDFormat fd_formats[] = {
1528 /* First entry is default format */
1529 /* 1.44 MB 3"1/2 floppy disks */
1530 { FDRIVE_DRV_144, 18, 80, 1, },
1531 { FDRIVE_DRV_144, 20, 80, 1, },
1532 { FDRIVE_DRV_144, 21, 80, 1, },
1533 { FDRIVE_DRV_144, 21, 82, 1, },
1534 { FDRIVE_DRV_144, 21, 83, 1, },
1535 { FDRIVE_DRV_144, 22, 80, 1, },
1536 { FDRIVE_DRV_144, 23, 80, 1, },
1537 { FDRIVE_DRV_144, 24, 80, 1, },
1538 /* 2.88 MB 3"1/2 floppy disks */
1539 { FDRIVE_DRV_288, 36, 80, 1, },
1540 { FDRIVE_DRV_288, 39, 80, 1, },
1541 { FDRIVE_DRV_288, 40, 80, 1, },
1542 { FDRIVE_DRV_288, 44, 80, 1, },
1543 { FDRIVE_DRV_288, 48, 80, 1, },
1544 /* 720 kB 3"1/2 floppy disks */
1545 { FDRIVE_DRV_144, 9, 80, 1, },
1546 { FDRIVE_DRV_144, 10, 80, 1, },
1547 { FDRIVE_DRV_144, 10, 82, 1, },
1548 { FDRIVE_DRV_144, 10, 83, 1, },
1549 { FDRIVE_DRV_144, 13, 80, 1, },
1550 { FDRIVE_DRV_144, 14, 80, 1, },
1551 /* 1.2 MB 5"1/4 floppy disks */
1552 { FDRIVE_DRV_120, 15, 80, 1, },
1553 { FDRIVE_DRV_120, 18, 80, 1, },
1554 { FDRIVE_DRV_120, 18, 82, 1, },
1555 { FDRIVE_DRV_120, 18, 83, 1, },
1556 { FDRIVE_DRV_120, 20, 80, 1, },
1557 /* 720 kB 5"1/4 floppy disks */
1558 { FDRIVE_DRV_120, 9, 80, 1, },
1559 { FDRIVE_DRV_120, 11, 80, 1, },
1560 /* 360 kB 5"1/4 floppy disks */
1561 { FDRIVE_DRV_120, 9, 40, 1, },
1562 { FDRIVE_DRV_120, 9, 40, 0, },
1563 { FDRIVE_DRV_120, 10, 41, 1, },
1564 { FDRIVE_DRV_120, 10, 42, 1, },
1565 /* 320 kB 5"1/4 floppy disks */
1566 { FDRIVE_DRV_120, 8, 40, 1, },
1567 { FDRIVE_DRV_120, 8, 40, 0, },
1568 /* 360 kB must match 5"1/4 better than 3"1/2... */
1569 { FDRIVE_DRV_144, 9, 80, 0, },
1570 /* end */
1571 { FDRIVE_DRV_NONE, -1, -1, 0, },
1572 };
1573
1574 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
1575 int *max_track, int *last_sect,
1576 FDriveType drive_in, FDriveType *drive)
1577 {
1578 const FDFormat *parse;
1579 uint64_t nb_sectors, size;
1580 int i, first_match, match;
1581
1582 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
1583 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
1584 /* User defined disk */
1585 } else {
1586 bdrv_get_geometry(bs, &nb_sectors);
1587 match = -1;
1588 first_match = -1;
1589 for (i = 0; ; i++) {
1590 parse = &fd_formats[i];
1591 if (parse->drive == FDRIVE_DRV_NONE) {
1592 break;
1593 }
1594 if (drive_in == parse->drive ||
1595 drive_in == FDRIVE_DRV_NONE) {
1596 size = (parse->max_head + 1) * parse->max_track *
1597 parse->last_sect;
1598 if (nb_sectors == size) {
1599 match = i;
1600 break;
1601 }
1602 if (first_match == -1) {
1603 first_match = i;
1604 }
1605 }
1606 }
1607 if (match == -1) {
1608 if (first_match == -1) {
1609 match = 1;
1610 } else {
1611 match = first_match;
1612 }
1613 parse = &fd_formats[match];
1614 }
1615 *nb_heads = parse->max_head + 1;
1616 *max_track = parse->max_track;
1617 *last_sect = parse->last_sect;
1618 *drive = parse->drive;
1619 }
1620 }
1621
1622 int bdrv_get_translation_hint(BlockDriverState *bs)
1623 {
1624 return bs->translation;
1625 }
1626
1627 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1628 BlockErrorAction on_write_error)
1629 {
1630 bs->on_read_error = on_read_error;
1631 bs->on_write_error = on_write_error;
1632 }
1633
1634 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1635 {
1636 return is_read ? bs->on_read_error : bs->on_write_error;
1637 }
1638
1639 int bdrv_is_read_only(BlockDriverState *bs)
1640 {
1641 return bs->read_only;
1642 }
1643
1644 int bdrv_is_sg(BlockDriverState *bs)
1645 {
1646 return bs->sg;
1647 }
1648
1649 int bdrv_enable_write_cache(BlockDriverState *bs)
1650 {
1651 return bs->enable_write_cache;
1652 }
1653
1654 int bdrv_is_encrypted(BlockDriverState *bs)
1655 {
1656 if (bs->backing_hd && bs->backing_hd->encrypted)
1657 return 1;
1658 return bs->encrypted;
1659 }
1660
1661 int bdrv_key_required(BlockDriverState *bs)
1662 {
1663 BlockDriverState *backing_hd = bs->backing_hd;
1664
1665 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1666 return 1;
1667 return (bs->encrypted && !bs->valid_key);
1668 }
1669
1670 int bdrv_set_key(BlockDriverState *bs, const char *key)
1671 {
1672 int ret;
1673 if (bs->backing_hd && bs->backing_hd->encrypted) {
1674 ret = bdrv_set_key(bs->backing_hd, key);
1675 if (ret < 0)
1676 return ret;
1677 if (!bs->encrypted)
1678 return 0;
1679 }
1680 if (!bs->encrypted) {
1681 return -EINVAL;
1682 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1683 return -ENOMEDIUM;
1684 }
1685 ret = bs->drv->bdrv_set_key(bs, key);
1686 if (ret < 0) {
1687 bs->valid_key = 0;
1688 } else if (!bs->valid_key) {
1689 bs->valid_key = 1;
1690 /* call the change callback now, we skipped it on open */
1691 bdrv_dev_change_media_cb(bs, true);
1692 }
1693 return ret;
1694 }
1695
1696 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1697 {
1698 if (!bs->drv) {
1699 buf[0] = '\0';
1700 } else {
1701 pstrcpy(buf, buf_size, bs->drv->format_name);
1702 }
1703 }
1704
1705 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1706 void *opaque)
1707 {
1708 BlockDriver *drv;
1709
1710 QLIST_FOREACH(drv, &bdrv_drivers, list) {
1711 it(opaque, drv->format_name);
1712 }
1713 }
1714
1715 BlockDriverState *bdrv_find(const char *name)
1716 {
1717 BlockDriverState *bs;
1718
1719 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1720 if (!strcmp(name, bs->device_name)) {
1721 return bs;
1722 }
1723 }
1724 return NULL;
1725 }
1726
1727 BlockDriverState *bdrv_next(BlockDriverState *bs)
1728 {
1729 if (!bs) {
1730 return QTAILQ_FIRST(&bdrv_states);
1731 }
1732 return QTAILQ_NEXT(bs, list);
1733 }
1734
1735 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1736 {
1737 BlockDriverState *bs;
1738
1739 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1740 it(opaque, bs);
1741 }
1742 }
1743
1744 const char *bdrv_get_device_name(BlockDriverState *bs)
1745 {
1746 return bs->device_name;
1747 }
1748
1749 void bdrv_flush_all(void)
1750 {
1751 BlockDriverState *bs;
1752
1753 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1754 if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
1755 bdrv_flush(bs);
1756 }
1757 }
1758 }
1759
1760 int bdrv_has_zero_init(BlockDriverState *bs)
1761 {
1762 assert(bs->drv);
1763
1764 if (bs->drv->bdrv_has_zero_init) {
1765 return bs->drv->bdrv_has_zero_init(bs);
1766 }
1767
1768 return 1;
1769 }
1770
1771 /*
1772 * Returns true iff the specified sector is present in the disk image. Drivers
1773 * not implementing the functionality are assumed to not support backing files,
1774 * hence all their sectors are reported as allocated.
1775 *
1776 * 'pnum' is set to the number of sectors (including and immediately following
1777 * the specified sector) that are known to be in the same
1778 * allocated/unallocated state.
1779 *
1780 * 'nb_sectors' is the max value 'pnum' should be set to.
1781 */
1782 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1783 int *pnum)
1784 {
1785 int64_t n;
1786 if (!bs->drv->bdrv_is_allocated) {
1787 if (sector_num >= bs->total_sectors) {
1788 *pnum = 0;
1789 return 0;
1790 }
1791 n = bs->total_sectors - sector_num;
1792 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1793 return 1;
1794 }
1795 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1796 }
1797
1798 void bdrv_mon_event(const BlockDriverState *bdrv,
1799 BlockMonEventAction action, int is_read)
1800 {
1801 QObject *data;
1802 const char *action_str;
1803
1804 switch (action) {
1805 case BDRV_ACTION_REPORT:
1806 action_str = "report";
1807 break;
1808 case BDRV_ACTION_IGNORE:
1809 action_str = "ignore";
1810 break;
1811 case BDRV_ACTION_STOP:
1812 action_str = "stop";
1813 break;
1814 default:
1815 abort();
1816 }
1817
1818 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1819 bdrv->device_name,
1820 action_str,
1821 is_read ? "read" : "write");
1822 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1823
1824 qobject_decref(data);
1825 }
1826
1827 static void bdrv_print_dict(QObject *obj, void *opaque)
1828 {
1829 QDict *bs_dict;
1830 Monitor *mon = opaque;
1831
1832 bs_dict = qobject_to_qdict(obj);
1833
1834 monitor_printf(mon, "%s: removable=%d",
1835 qdict_get_str(bs_dict, "device"),
1836 qdict_get_bool(bs_dict, "removable"));
1837
1838 if (qdict_get_bool(bs_dict, "removable")) {
1839 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1840 monitor_printf(mon, " tray-open=%d",
1841 qdict_get_bool(bs_dict, "tray-open"));
1842 }
1843
1844 if (qdict_haskey(bs_dict, "io-status")) {
1845 monitor_printf(mon, " io-status=%s", qdict_get_str(bs_dict, "io-status"));
1846 }
1847
1848 if (qdict_haskey(bs_dict, "inserted")) {
1849 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1850
1851 monitor_printf(mon, " file=");
1852 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1853 if (qdict_haskey(qdict, "backing_file")) {
1854 monitor_printf(mon, " backing_file=");
1855 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1856 }
1857 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1858 qdict_get_bool(qdict, "ro"),
1859 qdict_get_str(qdict, "drv"),
1860 qdict_get_bool(qdict, "encrypted"));
1861 } else {
1862 monitor_printf(mon, " [not inserted]");
1863 }
1864
1865 monitor_printf(mon, "\n");
1866 }
1867
1868 void bdrv_info_print(Monitor *mon, const QObject *data)
1869 {
1870 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1871 }
1872
1873 static const char *const io_status_name[BDRV_IOS_MAX] = {
1874 [BDRV_IOS_OK] = "ok",
1875 [BDRV_IOS_FAILED] = "failed",
1876 [BDRV_IOS_ENOSPC] = "nospace",
1877 };
1878
1879 void bdrv_info(Monitor *mon, QObject **ret_data)
1880 {
1881 QList *bs_list;
1882 BlockDriverState *bs;
1883
1884 bs_list = qlist_new();
1885
1886 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1887 QObject *bs_obj;
1888 QDict *bs_dict;
1889
1890 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': 'unknown', "
1891 "'removable': %i, 'locked': %i }",
1892 bs->device_name,
1893 bdrv_dev_has_removable_media(bs),
1894 bdrv_dev_is_medium_locked(bs));
1895 bs_dict = qobject_to_qdict(bs_obj);
1896
1897 if (bdrv_dev_has_removable_media(bs)) {
1898 qdict_put(bs_dict, "tray-open",
1899 qbool_from_int(bdrv_dev_is_tray_open(bs)));
1900 }
1901
1902 if (bdrv_iostatus_is_enabled(bs)) {
1903 qdict_put(bs_dict, "io-status",
1904 qstring_from_str(io_status_name[bs->iostatus]));
1905 }
1906
1907 if (bs->drv) {
1908 QObject *obj;
1909
1910 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1911 "'encrypted': %i }",
1912 bs->filename, bs->read_only,
1913 bs->drv->format_name,
1914 bdrv_is_encrypted(bs));
1915 if (bs->backing_file[0] != '\0') {
1916 QDict *qdict = qobject_to_qdict(obj);
1917 qdict_put(qdict, "backing_file",
1918 qstring_from_str(bs->backing_file));
1919 }
1920
1921 qdict_put_obj(bs_dict, "inserted", obj);
1922 }
1923 qlist_append_obj(bs_list, bs_obj);
1924 }
1925
1926 *ret_data = QOBJECT(bs_list);
1927 }
1928
1929 static void bdrv_stats_iter(QObject *data, void *opaque)
1930 {
1931 QDict *qdict;
1932 Monitor *mon = opaque;
1933
1934 qdict = qobject_to_qdict(data);
1935 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1936
1937 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1938 monitor_printf(mon, " rd_bytes=%" PRId64
1939 " wr_bytes=%" PRId64
1940 " rd_operations=%" PRId64
1941 " wr_operations=%" PRId64
1942 " flush_operations=%" PRId64
1943 " wr_total_time_ns=%" PRId64
1944 " rd_total_time_ns=%" PRId64
1945 " flush_total_time_ns=%" PRId64
1946 "\n",
1947 qdict_get_int(qdict, "rd_bytes"),
1948 qdict_get_int(qdict, "wr_bytes"),
1949 qdict_get_int(qdict, "rd_operations"),
1950 qdict_get_int(qdict, "wr_operations"),
1951 qdict_get_int(qdict, "flush_operations"),
1952 qdict_get_int(qdict, "wr_total_time_ns"),
1953 qdict_get_int(qdict, "rd_total_time_ns"),
1954 qdict_get_int(qdict, "flush_total_time_ns"));
1955 }
1956
1957 void bdrv_stats_print(Monitor *mon, const QObject *data)
1958 {
1959 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1960 }
1961
1962 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1963 {
1964 QObject *res;
1965 QDict *dict;
1966
1967 res = qobject_from_jsonf("{ 'stats': {"
1968 "'rd_bytes': %" PRId64 ","
1969 "'wr_bytes': %" PRId64 ","
1970 "'rd_operations': %" PRId64 ","
1971 "'wr_operations': %" PRId64 ","
1972 "'wr_highest_offset': %" PRId64 ","
1973 "'flush_operations': %" PRId64 ","
1974 "'wr_total_time_ns': %" PRId64 ","
1975 "'rd_total_time_ns': %" PRId64 ","
1976 "'flush_total_time_ns': %" PRId64
1977 "} }",
1978 bs->nr_bytes[BDRV_ACCT_READ],
1979 bs->nr_bytes[BDRV_ACCT_WRITE],
1980 bs->nr_ops[BDRV_ACCT_READ],
1981 bs->nr_ops[BDRV_ACCT_WRITE],
1982 bs->wr_highest_sector *
1983 (uint64_t)BDRV_SECTOR_SIZE,
1984 bs->nr_ops[BDRV_ACCT_FLUSH],
1985 bs->total_time_ns[BDRV_ACCT_WRITE],
1986 bs->total_time_ns[BDRV_ACCT_READ],
1987 bs->total_time_ns[BDRV_ACCT_FLUSH]);
1988 dict = qobject_to_qdict(res);
1989
1990 if (*bs->device_name) {
1991 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1992 }
1993
1994 if (bs->file) {
1995 QObject *parent = bdrv_info_stats_bs(bs->file);
1996 qdict_put_obj(dict, "parent", parent);
1997 }
1998
1999 return res;
2000 }
2001
2002 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
2003 {
2004 QObject *obj;
2005 QList *devices;
2006 BlockDriverState *bs;
2007
2008 devices = qlist_new();
2009
2010 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2011 obj = bdrv_info_stats_bs(bs);
2012 qlist_append_obj(devices, obj);
2013 }
2014
2015 *ret_data = QOBJECT(devices);
2016 }
2017
2018 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2019 {
2020 if (bs->backing_hd && bs->backing_hd->encrypted)
2021 return bs->backing_file;
2022 else if (bs->encrypted)
2023 return bs->filename;
2024 else
2025 return NULL;
2026 }
2027
2028 void bdrv_get_backing_filename(BlockDriverState *bs,
2029 char *filename, int filename_size)
2030 {
2031 if (!bs->backing_file) {
2032 pstrcpy(filename, filename_size, "");
2033 } else {
2034 pstrcpy(filename, filename_size, bs->backing_file);
2035 }
2036 }
2037
2038 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2039 const uint8_t *buf, int nb_sectors)
2040 {
2041 BlockDriver *drv = bs->drv;
2042 if (!drv)
2043 return -ENOMEDIUM;
2044 if (!drv->bdrv_write_compressed)
2045 return -ENOTSUP;
2046 if (bdrv_check_request(bs, sector_num, nb_sectors))
2047 return -EIO;
2048
2049 if (bs->dirty_bitmap) {
2050 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2051 }
2052
2053 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2054 }
2055
2056 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2057 {
2058 BlockDriver *drv = bs->drv;
2059 if (!drv)
2060 return -ENOMEDIUM;
2061 if (!drv->bdrv_get_info)
2062 return -ENOTSUP;
2063 memset(bdi, 0, sizeof(*bdi));
2064 return drv->bdrv_get_info(bs, bdi);
2065 }
2066
2067 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2068 int64_t pos, int size)
2069 {
2070 BlockDriver *drv = bs->drv;
2071 if (!drv)
2072 return -ENOMEDIUM;
2073 if (drv->bdrv_save_vmstate)
2074 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2075 if (bs->file)
2076 return bdrv_save_vmstate(bs->file, buf, pos, size);
2077 return -ENOTSUP;
2078 }
2079
2080 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2081 int64_t pos, int size)
2082 {
2083 BlockDriver *drv = bs->drv;
2084 if (!drv)
2085 return -ENOMEDIUM;
2086 if (drv->bdrv_load_vmstate)
2087 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2088 if (bs->file)
2089 return bdrv_load_vmstate(bs->file, buf, pos, size);
2090 return -ENOTSUP;
2091 }
2092
2093 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2094 {
2095 BlockDriver *drv = bs->drv;
2096
2097 if (!drv || !drv->bdrv_debug_event) {
2098 return;
2099 }
2100
2101 return drv->bdrv_debug_event(bs, event);
2102
2103 }
2104
2105 /**************************************************************/
2106 /* handling of snapshots */
2107
2108 int bdrv_can_snapshot(BlockDriverState *bs)
2109 {
2110 BlockDriver *drv = bs->drv;
2111 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2112 return 0;
2113 }
2114
2115 if (!drv->bdrv_snapshot_create) {
2116 if (bs->file != NULL) {
2117 return bdrv_can_snapshot(bs->file);
2118 }
2119 return 0;
2120 }
2121
2122 return 1;
2123 }
2124
2125 int bdrv_is_snapshot(BlockDriverState *bs)
2126 {
2127 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2128 }
2129
2130 BlockDriverState *bdrv_snapshots(void)
2131 {
2132 BlockDriverState *bs;
2133
2134 if (bs_snapshots) {
2135 return bs_snapshots;
2136 }
2137
2138 bs = NULL;
2139 while ((bs = bdrv_next(bs))) {
2140 if (bdrv_can_snapshot(bs)) {
2141 bs_snapshots = bs;
2142 return bs;
2143 }
2144 }
2145 return NULL;
2146 }
2147
2148 int bdrv_snapshot_create(BlockDriverState *bs,
2149 QEMUSnapshotInfo *sn_info)
2150 {
2151 BlockDriver *drv = bs->drv;
2152 if (!drv)
2153 return -ENOMEDIUM;
2154 if (drv->bdrv_snapshot_create)
2155 return drv->bdrv_snapshot_create(bs, sn_info);
2156 if (bs->file)
2157 return bdrv_snapshot_create(bs->file, sn_info);
2158 return -ENOTSUP;
2159 }
2160
2161 int bdrv_snapshot_goto(BlockDriverState *bs,
2162 const char *snapshot_id)
2163 {
2164 BlockDriver *drv = bs->drv;
2165 int ret, open_ret;
2166
2167 if (!drv)
2168 return -ENOMEDIUM;
2169 if (drv->bdrv_snapshot_goto)
2170 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2171
2172 if (bs->file) {
2173 drv->bdrv_close(bs);
2174 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2175 open_ret = drv->bdrv_open(bs, bs->open_flags);
2176 if (open_ret < 0) {
2177 bdrv_delete(bs->file);
2178 bs->drv = NULL;
2179 return open_ret;
2180 }
2181 return ret;
2182 }
2183
2184 return -ENOTSUP;
2185 }
2186
2187 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2188 {
2189 BlockDriver *drv = bs->drv;
2190 if (!drv)
2191 return -ENOMEDIUM;
2192 if (drv->bdrv_snapshot_delete)
2193 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2194 if (bs->file)
2195 return bdrv_snapshot_delete(bs->file, snapshot_id);
2196 return -ENOTSUP;
2197 }
2198
2199 int bdrv_snapshot_list(BlockDriverState *bs,
2200 QEMUSnapshotInfo **psn_info)
2201 {
2202 BlockDriver *drv = bs->drv;
2203 if (!drv)
2204 return -ENOMEDIUM;
2205 if (drv->bdrv_snapshot_list)
2206 return drv->bdrv_snapshot_list(bs, psn_info);
2207 if (bs->file)
2208 return bdrv_snapshot_list(bs->file, psn_info);
2209 return -ENOTSUP;
2210 }
2211
2212 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2213 const char *snapshot_name)
2214 {
2215 BlockDriver *drv = bs->drv;
2216 if (!drv) {
2217 return -ENOMEDIUM;
2218 }
2219 if (!bs->read_only) {
2220 return -EINVAL;
2221 }
2222 if (drv->bdrv_snapshot_load_tmp) {
2223 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2224 }
2225 return -ENOTSUP;
2226 }
2227
2228 #define NB_SUFFIXES 4
2229
2230 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2231 {
2232 static const char suffixes[NB_SUFFIXES] = "KMGT";
2233 int64_t base;
2234 int i;
2235
2236 if (size <= 999) {
2237 snprintf(buf, buf_size, "%" PRId64, size);
2238 } else {
2239 base = 1024;
2240 for(i = 0; i < NB_SUFFIXES; i++) {
2241 if (size < (10 * base)) {
2242 snprintf(buf, buf_size, "%0.1f%c",
2243 (double)size / base,
2244 suffixes[i]);
2245 break;
2246 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2247 snprintf(buf, buf_size, "%" PRId64 "%c",
2248 ((size + (base >> 1)) / base),
2249 suffixes[i]);
2250 break;
2251 }
2252 base = base * 1024;
2253 }
2254 }
2255 return buf;
2256 }
2257
2258 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2259 {
2260 char buf1[128], date_buf[128], clock_buf[128];
2261 #ifdef _WIN32
2262 struct tm *ptm;
2263 #else
2264 struct tm tm;
2265 #endif
2266 time_t ti;
2267 int64_t secs;
2268
2269 if (!sn) {
2270 snprintf(buf, buf_size,
2271 "%-10s%-20s%7s%20s%15s",
2272 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2273 } else {
2274 ti = sn->date_sec;
2275 #ifdef _WIN32
2276 ptm = localtime(&ti);
2277 strftime(date_buf, sizeof(date_buf),
2278 "%Y-%m-%d %H:%M:%S", ptm);
2279 #else
2280 localtime_r(&ti, &tm);
2281 strftime(date_buf, sizeof(date_buf),
2282 "%Y-%m-%d %H:%M:%S", &tm);
2283 #endif
2284 secs = sn->vm_clock_nsec / 1000000000;
2285 snprintf(clock_buf, sizeof(clock_buf),
2286 "%02d:%02d:%02d.%03d",
2287 (int)(secs / 3600),
2288 (int)((secs / 60) % 60),
2289 (int)(secs % 60),
2290 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2291 snprintf(buf, buf_size,
2292 "%-10s%-20s%7s%20s%15s",
2293 sn->id_str, sn->name,
2294 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2295 date_buf,
2296 clock_buf);
2297 }
2298 return buf;
2299 }
2300
2301 /**************************************************************/
2302 /* async I/Os */
2303
2304 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2305 QEMUIOVector *qiov, int nb_sectors,
2306 BlockDriverCompletionFunc *cb, void *opaque)
2307 {
2308 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2309
2310 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2311 cb, opaque, false);
2312 }
2313
2314 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2315 QEMUIOVector *qiov, int nb_sectors,
2316 BlockDriverCompletionFunc *cb, void *opaque)
2317 {
2318 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2319
2320 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2321 cb, opaque, true);
2322 }
2323
2324
2325 typedef struct MultiwriteCB {
2326 int error;
2327 int num_requests;
2328 int num_callbacks;
2329 struct {
2330 BlockDriverCompletionFunc *cb;
2331 void *opaque;
2332 QEMUIOVector *free_qiov;
2333 void *free_buf;
2334 } callbacks[];
2335 } MultiwriteCB;
2336
2337 static void multiwrite_user_cb(MultiwriteCB *mcb)
2338 {
2339 int i;
2340
2341 for (i = 0; i < mcb->num_callbacks; i++) {
2342 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2343 if (mcb->callbacks[i].free_qiov) {
2344 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2345 }
2346 g_free(mcb->callbacks[i].free_qiov);
2347 qemu_vfree(mcb->callbacks[i].free_buf);
2348 }
2349 }
2350
2351 static void multiwrite_cb(void *opaque, int ret)
2352 {
2353 MultiwriteCB *mcb = opaque;
2354
2355 trace_multiwrite_cb(mcb, ret);
2356
2357 if (ret < 0 && !mcb->error) {
2358 mcb->error = ret;
2359 }
2360
2361 mcb->num_requests--;
2362 if (mcb->num_requests == 0) {
2363 multiwrite_user_cb(mcb);
2364 g_free(mcb);
2365 }
2366 }
2367
2368 static int multiwrite_req_compare(const void *a, const void *b)
2369 {
2370 const BlockRequest *req1 = a, *req2 = b;
2371
2372 /*
2373 * Note that we can't simply subtract req2->sector from req1->sector
2374 * here as that could overflow the return value.
2375 */
2376 if (req1->sector > req2->sector) {
2377 return 1;
2378 } else if (req1->sector < req2->sector) {
2379 return -1;
2380 } else {
2381 return 0;
2382 }
2383 }
2384
2385 /*
2386 * Takes a bunch of requests and tries to merge them. Returns the number of
2387 * requests that remain after merging.
2388 */
2389 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2390 int num_reqs, MultiwriteCB *mcb)
2391 {
2392 int i, outidx;
2393
2394 // Sort requests by start sector
2395 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2396
2397 // Check if adjacent requests touch the same clusters. If so, combine them,
2398 // filling up gaps with zero sectors.
2399 outidx = 0;
2400 for (i = 1; i < num_reqs; i++) {
2401 int merge = 0;
2402 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2403
2404 // This handles the cases that are valid for all block drivers, namely
2405 // exactly sequential writes and overlapping writes.
2406 if (reqs[i].sector <= oldreq_last) {
2407 merge = 1;
2408 }
2409
2410 // The block driver may decide that it makes sense to combine requests
2411 // even if there is a gap of some sectors between them. In this case,
2412 // the gap is filled with zeros (therefore only applicable for yet
2413 // unused space in format like qcow2).
2414 if (!merge && bs->drv->bdrv_merge_requests) {
2415 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2416 }
2417
2418 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2419 merge = 0;
2420 }
2421
2422 if (merge) {
2423 size_t size;
2424 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
2425 qemu_iovec_init(qiov,
2426 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2427
2428 // Add the first request to the merged one. If the requests are
2429 // overlapping, drop the last sectors of the first request.
2430 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2431 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2432
2433 // We might need to add some zeros between the two requests
2434 if (reqs[i].sector > oldreq_last) {
2435 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2436 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2437 memset(buf, 0, zero_bytes);
2438 qemu_iovec_add(qiov, buf, zero_bytes);
2439 mcb->callbacks[i].free_buf = buf;
2440 }
2441
2442 // Add the second request
2443 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2444
2445 reqs[outidx].nb_sectors = qiov->size >> 9;
2446 reqs[outidx].qiov = qiov;
2447
2448 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2449 } else {
2450 outidx++;
2451 reqs[outidx].sector = reqs[i].sector;
2452 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2453 reqs[outidx].qiov = reqs[i].qiov;
2454 }
2455 }
2456
2457 return outidx + 1;
2458 }
2459
2460 /*
2461 * Submit multiple AIO write requests at once.
2462 *
2463 * On success, the function returns 0 and all requests in the reqs array have
2464 * been submitted. In error case this function returns -1, and any of the
2465 * requests may or may not be submitted yet. In particular, this means that the
2466 * callback will be called for some of the requests, for others it won't. The
2467 * caller must check the error field of the BlockRequest to wait for the right
2468 * callbacks (if error != 0, no callback will be called).
2469 *
2470 * The implementation may modify the contents of the reqs array, e.g. to merge
2471 * requests. However, the fields opaque and error are left unmodified as they
2472 * are used to signal failure for a single request to the caller.
2473 */
2474 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2475 {
2476 BlockDriverAIOCB *acb;
2477 MultiwriteCB *mcb;
2478 int i;
2479
2480 /* don't submit writes if we don't have a medium */
2481 if (bs->drv == NULL) {
2482 for (i = 0; i < num_reqs; i++) {
2483 reqs[i].error = -ENOMEDIUM;
2484 }
2485 return -1;
2486 }
2487
2488 if (num_reqs == 0) {
2489 return 0;
2490 }
2491
2492 // Create MultiwriteCB structure
2493 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2494 mcb->num_requests = 0;
2495 mcb->num_callbacks = num_reqs;
2496
2497 for (i = 0; i < num_reqs; i++) {
2498 mcb->callbacks[i].cb = reqs[i].cb;
2499 mcb->callbacks[i].opaque = reqs[i].opaque;
2500 }
2501
2502 // Check for mergable requests
2503 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2504
2505 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2506
2507 /*
2508 * Run the aio requests. As soon as one request can't be submitted
2509 * successfully, fail all requests that are not yet submitted (we must
2510 * return failure for all requests anyway)
2511 *
2512 * num_requests cannot be set to the right value immediately: If
2513 * bdrv_aio_writev fails for some request, num_requests would be too high
2514 * and therefore multiwrite_cb() would never recognize the multiwrite
2515 * request as completed. We also cannot use the loop variable i to set it
2516 * when the first request fails because the callback may already have been
2517 * called for previously submitted requests. Thus, num_requests must be
2518 * incremented for each request that is submitted.
2519 *
2520 * The problem that callbacks may be called early also means that we need
2521 * to take care that num_requests doesn't become 0 before all requests are
2522 * submitted - multiwrite_cb() would consider the multiwrite request
2523 * completed. A dummy request that is "completed" by a manual call to
2524 * multiwrite_cb() takes care of this.
2525 */
2526 mcb->num_requests = 1;
2527
2528 // Run the aio requests
2529 for (i = 0; i < num_reqs; i++) {
2530 mcb->num_requests++;
2531 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2532 reqs[i].nb_sectors, multiwrite_cb, mcb);
2533
2534 if (acb == NULL) {
2535 // We can only fail the whole thing if no request has been
2536 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2537 // complete and report the error in the callback.
2538 if (i == 0) {
2539 trace_bdrv_aio_multiwrite_earlyfail(mcb);
2540 goto fail;
2541 } else {
2542 trace_bdrv_aio_multiwrite_latefail(mcb, i);
2543 multiwrite_cb(mcb, -EIO);
2544 break;
2545 }
2546 }
2547 }
2548
2549 /* Complete the dummy request */
2550 multiwrite_cb(mcb, 0);
2551
2552 return 0;
2553
2554 fail:
2555 for (i = 0; i < mcb->num_callbacks; i++) {
2556 reqs[i].error = -EIO;
2557 }
2558 g_free(mcb);
2559 return -1;
2560 }
2561
2562 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2563 {
2564 acb->pool->cancel(acb);
2565 }
2566
2567
2568 /**************************************************************/
2569 /* async block device emulation */
2570
2571 typedef struct BlockDriverAIOCBSync {
2572 BlockDriverAIOCB common;
2573 QEMUBH *bh;
2574 int ret;
2575 /* vector translation state */
2576 QEMUIOVector *qiov;
2577 uint8_t *bounce;
2578 int is_write;
2579 } BlockDriverAIOCBSync;
2580
2581 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2582 {
2583 BlockDriverAIOCBSync *acb =
2584 container_of(blockacb, BlockDriverAIOCBSync, common);
2585 qemu_bh_delete(acb->bh);
2586 acb->bh = NULL;
2587 qemu_aio_release(acb);
2588 }
2589
2590 static AIOPool bdrv_em_aio_pool = {
2591 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2592 .cancel = bdrv_aio_cancel_em,
2593 };
2594
2595 static void bdrv_aio_bh_cb(void *opaque)
2596 {
2597 BlockDriverAIOCBSync *acb = opaque;
2598
2599 if (!acb->is_write)
2600 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2601 qemu_vfree(acb->bounce);
2602 acb->common.cb(acb->common.opaque, acb->ret);
2603 qemu_bh_delete(acb->bh);
2604 acb->bh = NULL;
2605 qemu_aio_release(acb);
2606 }
2607
2608 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2609 int64_t sector_num,
2610 QEMUIOVector *qiov,
2611 int nb_sectors,
2612 BlockDriverCompletionFunc *cb,
2613 void *opaque,
2614 int is_write)
2615
2616 {
2617 BlockDriverAIOCBSync *acb;
2618
2619 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2620 acb->is_write = is_write;
2621 acb->qiov = qiov;
2622 acb->bounce = qemu_blockalign(bs, qiov->size);
2623
2624 if (!acb->bh)
2625 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2626
2627 if (is_write) {
2628 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2629 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2630 } else {
2631 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2632 }
2633
2634 qemu_bh_schedule(acb->bh);
2635
2636 return &acb->common;
2637 }
2638
2639 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2640 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2641 BlockDriverCompletionFunc *cb, void *opaque)
2642 {
2643 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2644 }
2645
2646 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2647 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2648 BlockDriverCompletionFunc *cb, void *opaque)
2649 {
2650 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2651 }
2652
2653
2654 typedef struct BlockDriverAIOCBCoroutine {
2655 BlockDriverAIOCB common;
2656 BlockRequest req;
2657 bool is_write;
2658 QEMUBH* bh;
2659 } BlockDriverAIOCBCoroutine;
2660
2661 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
2662 {
2663 qemu_aio_flush();
2664 }
2665
2666 static AIOPool bdrv_em_co_aio_pool = {
2667 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
2668 .cancel = bdrv_aio_co_cancel_em,
2669 };
2670
2671 static void bdrv_co_em_bh(void *opaque)
2672 {
2673 BlockDriverAIOCBCoroutine *acb = opaque;
2674
2675 acb->common.cb(acb->common.opaque, acb->req.error);
2676 qemu_bh_delete(acb->bh);
2677 qemu_aio_release(acb);
2678 }
2679
2680 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2681 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2682 {
2683 BlockDriverAIOCBCoroutine *acb = opaque;
2684 BlockDriverState *bs = acb->common.bs;
2685
2686 if (!acb->is_write) {
2687 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
2688 acb->req.nb_sectors, acb->req.qiov);
2689 } else {
2690 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
2691 acb->req.nb_sectors, acb->req.qiov);
2692 }
2693
2694 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
2695 qemu_bh_schedule(acb->bh);
2696 }
2697
2698 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
2699 int64_t sector_num,
2700 QEMUIOVector *qiov,
2701 int nb_sectors,
2702 BlockDriverCompletionFunc *cb,
2703 void *opaque,
2704 bool is_write)
2705 {
2706 Coroutine *co;
2707 BlockDriverAIOCBCoroutine *acb;
2708
2709 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
2710 acb->req.sector = sector_num;
2711 acb->req.nb_sectors = nb_sectors;
2712 acb->req.qiov = qiov;
2713 acb->is_write = is_write;
2714
2715 co = qemu_coroutine_create(bdrv_co_do_rw);
2716 qemu_coroutine_enter(co, acb);
2717
2718 return &acb->common;
2719 }
2720
2721 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2722 {
2723 BlockDriverAIOCBCoroutine *acb = opaque;
2724 BlockDriverState *bs = acb->common.bs;
2725
2726 acb->req.error = bdrv_co_flush(bs);
2727 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
2728 qemu_bh_schedule(acb->bh);
2729 }
2730
2731 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2732 BlockDriverCompletionFunc *cb, void *opaque)
2733 {
2734 trace_bdrv_aio_flush(bs, opaque);
2735
2736 Coroutine *co;
2737 BlockDriverAIOCBCoroutine *acb;
2738
2739 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
2740 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
2741 qemu_coroutine_enter(co, acb);
2742
2743 return &acb->common;
2744 }
2745
2746 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
2747 {
2748 BlockDriverAIOCBCoroutine *acb = opaque;
2749 BlockDriverState *bs = acb->common.bs;
2750
2751 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
2752 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
2753 qemu_bh_schedule(acb->bh);
2754 }
2755
2756 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
2757 int64_t sector_num, int nb_sectors,
2758 BlockDriverCompletionFunc *cb, void *opaque)
2759 {
2760 Coroutine *co;
2761 BlockDriverAIOCBCoroutine *acb;
2762
2763 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
2764
2765 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
2766 acb->req.sector = sector_num;
2767 acb->req.nb_sectors = nb_sectors;
2768 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
2769 qemu_coroutine_enter(co, acb);
2770
2771 return &acb->common;
2772 }
2773
2774 void bdrv_init(void)
2775 {
2776 module_call_init(MODULE_INIT_BLOCK);
2777 }
2778
2779 void bdrv_init_with_whitelist(void)
2780 {
2781 use_bdrv_whitelist = 1;
2782 bdrv_init();
2783 }
2784
2785 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2786 BlockDriverCompletionFunc *cb, void *opaque)
2787 {
2788 BlockDriverAIOCB *acb;
2789
2790 if (pool->free_aiocb) {
2791 acb = pool->free_aiocb;
2792 pool->free_aiocb = acb->next;
2793 } else {
2794 acb = g_malloc0(pool->aiocb_size);
2795 acb->pool = pool;
2796 }
2797 acb->bs = bs;
2798 acb->cb = cb;
2799 acb->opaque = opaque;
2800 return acb;
2801 }
2802
2803 void qemu_aio_release(void *p)
2804 {
2805 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2806 AIOPool *pool = acb->pool;
2807 acb->next = pool->free_aiocb;
2808 pool->free_aiocb = acb;
2809 }
2810
2811 /**************************************************************/
2812 /* Coroutine block device emulation */
2813
2814 typedef struct CoroutineIOCompletion {
2815 Coroutine *coroutine;
2816 int ret;
2817 } CoroutineIOCompletion;
2818
2819 static void bdrv_co_io_em_complete(void *opaque, int ret)
2820 {
2821 CoroutineIOCompletion *co = opaque;
2822
2823 co->ret = ret;
2824 qemu_coroutine_enter(co->coroutine, NULL);
2825 }
2826
2827 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
2828 int nb_sectors, QEMUIOVector *iov,
2829 bool is_write)
2830 {
2831 CoroutineIOCompletion co = {
2832 .coroutine = qemu_coroutine_self(),
2833 };
2834 BlockDriverAIOCB *acb;
2835
2836 if (is_write) {
2837 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
2838 bdrv_co_io_em_complete, &co);
2839 } else {
2840 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
2841 bdrv_co_io_em_complete, &co);
2842 }
2843
2844 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
2845 if (!acb) {
2846 return -EIO;
2847 }
2848 qemu_coroutine_yield();
2849
2850 return co.ret;
2851 }
2852
2853 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
2854 int64_t sector_num, int nb_sectors,
2855 QEMUIOVector *iov)
2856 {
2857 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
2858 }
2859
2860 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
2861 int64_t sector_num, int nb_sectors,
2862 QEMUIOVector *iov)
2863 {
2864 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
2865 }
2866
2867 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2868 {
2869 RwCo *rwco = opaque;
2870
2871 rwco->ret = bdrv_co_flush(rwco->bs);
2872 }
2873
2874 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2875 {
2876 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2877 return 0;
2878 } else if (!bs->drv) {
2879 return 0;
2880 } else if (bs->drv->bdrv_co_flush) {
2881 return bs->drv->bdrv_co_flush(bs);
2882 } else if (bs->drv->bdrv_aio_flush) {
2883 BlockDriverAIOCB *acb;
2884 CoroutineIOCompletion co = {
2885 .coroutine = qemu_coroutine_self(),
2886 };
2887
2888 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2889 if (acb == NULL) {
2890 return -EIO;
2891 } else {
2892 qemu_coroutine_yield();
2893 return co.ret;
2894 }
2895 } else {
2896 /*
2897 * Some block drivers always operate in either writethrough or unsafe
2898 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2899 * know how the server works (because the behaviour is hardcoded or
2900 * depends on server-side configuration), so we can't ensure that
2901 * everything is safe on disk. Returning an error doesn't work because
2902 * that would break guests even if the server operates in writethrough
2903 * mode.
2904 *
2905 * Let's hope the user knows what he's doing.
2906 */
2907 return 0;
2908 }
2909 }
2910
2911 int bdrv_flush(BlockDriverState *bs)
2912 {
2913 Coroutine *co;
2914 RwCo rwco = {
2915 .bs = bs,
2916 .ret = NOT_DONE,
2917 };
2918
2919 if (qemu_in_coroutine()) {
2920 /* Fast-path if already in coroutine context */
2921 bdrv_flush_co_entry(&rwco);
2922 } else {
2923 co = qemu_coroutine_create(bdrv_flush_co_entry);
2924 qemu_coroutine_enter(co, &rwco);
2925 while (rwco.ret == NOT_DONE) {
2926 qemu_aio_wait();
2927 }
2928 }
2929
2930 return rwco.ret;
2931 }
2932
2933 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
2934 {
2935 RwCo *rwco = opaque;
2936
2937 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
2938 }
2939
2940 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
2941 int nb_sectors)
2942 {
2943 if (!bs->drv) {
2944 return -ENOMEDIUM;
2945 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
2946 return -EIO;
2947 } else if (bs->read_only) {
2948 return -EROFS;
2949 } else if (bs->drv->bdrv_co_discard) {
2950 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
2951 } else if (bs->drv->bdrv_aio_discard) {
2952 BlockDriverAIOCB *acb;
2953 CoroutineIOCompletion co = {
2954 .coroutine = qemu_coroutine_self(),
2955 };
2956
2957 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
2958 bdrv_co_io_em_complete, &co);
2959 if (acb == NULL) {
2960 return -EIO;
2961 } else {
2962 qemu_coroutine_yield();
2963 return co.ret;
2964 }
2965 } else {
2966 return 0;
2967 }
2968 }
2969
2970 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
2971 {
2972 Coroutine *co;
2973 RwCo rwco = {
2974 .bs = bs,
2975 .sector_num = sector_num,
2976 .nb_sectors = nb_sectors,
2977 .ret = NOT_DONE,
2978 };
2979
2980 if (qemu_in_coroutine()) {
2981 /* Fast-path if already in coroutine context */
2982 bdrv_discard_co_entry(&rwco);
2983 } else {
2984 co = qemu_coroutine_create(bdrv_discard_co_entry);
2985 qemu_coroutine_enter(co, &rwco);
2986 while (rwco.ret == NOT_DONE) {
2987 qemu_aio_wait();
2988 }
2989 }
2990
2991 return rwco.ret;
2992 }
2993
2994 /**************************************************************/
2995 /* removable device support */
2996
2997 /**
2998 * Return TRUE if the media is present
2999 */
3000 int bdrv_is_inserted(BlockDriverState *bs)
3001 {
3002 BlockDriver *drv = bs->drv;
3003
3004 if (!drv)
3005 return 0;
3006 if (!drv->bdrv_is_inserted)
3007 return 1;
3008 return drv->bdrv_is_inserted(bs);
3009 }
3010
3011 /**
3012 * Return whether the media changed since the last call to this
3013 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3014 */
3015 int bdrv_media_changed(BlockDriverState *bs)
3016 {
3017 BlockDriver *drv = bs->drv;
3018
3019 if (drv && drv->bdrv_media_changed) {
3020 return drv->bdrv_media_changed(bs);
3021 }
3022 return -ENOTSUP;
3023 }
3024
3025 /**
3026 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3027 */
3028 void bdrv_eject(BlockDriverState *bs, int eject_flag)
3029 {
3030 BlockDriver *drv = bs->drv;
3031
3032 if (drv && drv->bdrv_eject) {
3033 drv->bdrv_eject(bs, eject_flag);
3034 }
3035 }
3036
3037 /**
3038 * Lock or unlock the media (if it is locked, the user won't be able
3039 * to eject it manually).
3040 */
3041 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3042 {
3043 BlockDriver *drv = bs->drv;
3044
3045 trace_bdrv_lock_medium(bs, locked);
3046
3047 if (drv && drv->bdrv_lock_medium) {
3048 drv->bdrv_lock_medium(bs, locked);
3049 }
3050 }
3051
3052 /* needed for generic scsi interface */
3053
3054 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3055 {
3056 BlockDriver *drv = bs->drv;
3057
3058 if (drv && drv->bdrv_ioctl)
3059 return drv->bdrv_ioctl(bs, req, buf);
3060 return -ENOTSUP;
3061 }
3062
3063 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3064 unsigned long int req, void *buf,
3065 BlockDriverCompletionFunc *cb, void *opaque)
3066 {
3067 BlockDriver *drv = bs->drv;
3068
3069 if (drv && drv->bdrv_aio_ioctl)
3070 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3071 return NULL;
3072 }
3073
3074 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3075 {
3076 bs->buffer_alignment = align;
3077 }
3078
3079 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3080 {
3081 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3082 }
3083
3084 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3085 {
3086 int64_t bitmap_size;
3087
3088 bs->dirty_count = 0;
3089 if (enable) {
3090 if (!bs->dirty_bitmap) {
3091 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3092 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3093 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3094
3095 bs->dirty_bitmap = g_malloc0(bitmap_size);
3096 }
3097 } else {
3098 if (bs->dirty_bitmap) {
3099 g_free(bs->dirty_bitmap);
3100 bs->dirty_bitmap = NULL;
3101 }
3102 }
3103 }
3104
3105 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3106 {
3107 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3108
3109 if (bs->dirty_bitmap &&
3110 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3111 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3112 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3113 } else {
3114 return 0;
3115 }
3116 }
3117
3118 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3119 int nr_sectors)
3120 {
3121 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3122 }
3123
3124 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3125 {
3126 return bs->dirty_count;
3127 }
3128
3129 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3130 {
3131 assert(bs->in_use != in_use);
3132 bs->in_use = in_use;
3133 }
3134
3135 int bdrv_in_use(BlockDriverState *bs)
3136 {
3137 return bs->in_use;
3138 }
3139
3140 void bdrv_iostatus_enable(BlockDriverState *bs)
3141 {
3142 bs->iostatus = BDRV_IOS_OK;
3143 }
3144
3145 /* The I/O status is only enabled if the drive explicitly
3146 * enables it _and_ the VM is configured to stop on errors */
3147 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3148 {
3149 return (bs->iostatus != BDRV_IOS_INVAL &&
3150 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3151 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3152 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3153 }
3154
3155 void bdrv_iostatus_disable(BlockDriverState *bs)
3156 {
3157 bs->iostatus = BDRV_IOS_INVAL;
3158 }
3159
3160 void bdrv_iostatus_reset(BlockDriverState *bs)
3161 {
3162 if (bdrv_iostatus_is_enabled(bs)) {
3163 bs->iostatus = BDRV_IOS_OK;
3164 }
3165 }
3166
3167 /* XXX: Today this is set by device models because it makes the implementation
3168 quite simple. However, the block layer knows about the error, so it's
3169 possible to implement this without device models being involved */
3170 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3171 {
3172 if (bdrv_iostatus_is_enabled(bs) && bs->iostatus == BDRV_IOS_OK) {
3173 assert(error >= 0);
3174 bs->iostatus = error == ENOSPC ? BDRV_IOS_ENOSPC : BDRV_IOS_FAILED;
3175 }
3176 }
3177
3178 void
3179 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3180 enum BlockAcctType type)
3181 {
3182 assert(type < BDRV_MAX_IOTYPE);
3183
3184 cookie->bytes = bytes;
3185 cookie->start_time_ns = get_clock();
3186 cookie->type = type;
3187 }
3188
3189 void
3190 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3191 {
3192 assert(cookie->type < BDRV_MAX_IOTYPE);
3193
3194 bs->nr_bytes[cookie->type] += cookie->bytes;
3195 bs->nr_ops[cookie->type]++;
3196 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
3197 }
3198
3199 int bdrv_img_create(const char *filename, const char *fmt,
3200 const char *base_filename, const char *base_fmt,
3201 char *options, uint64_t img_size, int flags)
3202 {
3203 QEMUOptionParameter *param = NULL, *create_options = NULL;
3204 QEMUOptionParameter *backing_fmt, *backing_file, *size;
3205 BlockDriverState *bs = NULL;
3206 BlockDriver *drv, *proto_drv;
3207 BlockDriver *backing_drv = NULL;
3208 int ret = 0;
3209
3210 /* Find driver and parse its options */
3211 drv = bdrv_find_format(fmt);
3212 if (!drv) {
3213 error_report("Unknown file format '%s'", fmt);
3214 ret = -EINVAL;
3215 goto out;
3216 }
3217
3218 proto_drv = bdrv_find_protocol(filename);
3219 if (!proto_drv) {
3220 error_report("Unknown protocol '%s'", filename);
3221 ret = -EINVAL;
3222 goto out;
3223 }
3224
3225 create_options = append_option_parameters(create_options,
3226 drv->create_options);
3227 create_options = append_option_parameters(create_options,
3228 proto_drv->create_options);
3229
3230 /* Create parameter list with default values */
3231 param = parse_option_parameters("", create_options, param);
3232
3233 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3234
3235 /* Parse -o options */
3236 if (options) {
3237 param = parse_option_parameters(options, create_options, param);
3238 if (param == NULL) {
3239 error_report("Invalid options for file format '%s'.", fmt);
3240 ret = -EINVAL;
3241 goto out;
3242 }
3243 }
3244
3245 if (base_filename) {
3246 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3247 base_filename)) {
3248 error_report("Backing file not supported for file format '%s'",
3249 fmt);
3250 ret = -EINVAL;
3251 goto out;
3252 }
3253 }
3254
3255 if (base_fmt) {
3256 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
3257 error_report("Backing file format not supported for file "
3258 "format '%s'", fmt);
3259 ret = -EINVAL;
3260 goto out;
3261 }
3262 }
3263
3264 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
3265 if (backing_file && backing_file->value.s) {
3266 if (!strcmp(filename, backing_file->value.s)) {
3267 error_report("Error: Trying to create an image with the "
3268 "same filename as the backing file");
3269 ret = -EINVAL;
3270 goto out;
3271 }
3272 }
3273
3274 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
3275 if (backing_fmt && backing_fmt->value.s) {
3276 backing_drv = bdrv_find_format(backing_fmt->value.s);
3277 if (!backing_drv) {
3278 error_report("Unknown backing file format '%s'",
3279 backing_fmt->value.s);
3280 ret = -EINVAL;
3281 goto out;
3282 }
3283 }
3284
3285 // The size for the image must always be specified, with one exception:
3286 // If we are using a backing file, we can obtain the size from there
3287 size = get_option_parameter(param, BLOCK_OPT_SIZE);
3288 if (size && size->value.n == -1) {
3289 if (backing_file && backing_file->value.s) {
3290 uint64_t size;
3291 char buf[32];
3292
3293 bs = bdrv_new("");
3294
3295 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
3296 if (ret < 0) {
3297 error_report("Could not open '%s'", backing_file->value.s);
3298 goto out;
3299 }
3300 bdrv_get_geometry(bs, &size);
3301 size *= 512;
3302
3303 snprintf(buf, sizeof(buf), "%" PRId64, size);
3304 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3305 } else {
3306 error_report("Image creation needs a size parameter");
3307 ret = -EINVAL;
3308 goto out;
3309 }
3310 }
3311
3312 printf("Formatting '%s', fmt=%s ", filename, fmt);
3313 print_option_parameters(param);
3314 puts("");
3315
3316 ret = bdrv_create(drv, filename, param);
3317
3318 if (ret < 0) {
3319 if (ret == -ENOTSUP) {
3320 error_report("Formatting or formatting option not supported for "
3321 "file format '%s'", fmt);
3322 } else if (ret == -EFBIG) {
3323 error_report("The image size is too large for file format '%s'",
3324 fmt);
3325 } else {
3326 error_report("%s: error while creating %s: %s", filename, fmt,
3327 strerror(-ret));
3328 }
3329 }
3330
3331 out:
3332 free_option_parameters(create_options);
3333 free_option_parameters(param);
3334
3335 if (bs) {
3336 bdrv_delete(bs);
3337 }
3338
3339 return ret;
3340 }
QEmu