]> git.proxmox.com Git - mirror_qemu.git/blame - block.c
projects / mirror_qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
hmp: re-enable trace-file command
[mirror_qemu.git] / block.c
CommitLineData
fc01f7e7
FB		 1/*
2 * QEMU System Emulator block driver
5fafdf24 3 *
fc01f7e7 4 * Copyright (c) 2003 Fabrice Bellard
5fafdf24 5 *
fc01f7e7
FB		 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 */
3990d09a 24#include "config-host.h"
faf07963 25#include "qemu-common.h"
6d519a5f 26#include "trace.h"
376253ec 27#include "monitor.h"
ea2384d3 28#include "block_int.h"
5efa9d5a 29#include "module.h"
d15e5465 30#include "qemu-objects.h"
68485420 31#include "qemu-coroutine.h"
fc01f7e7 32
71e72a19 33#ifdef CONFIG_BSD
7674e7bf
FB		 34#include <sys/types.h>
35#include <sys/stat.h>
36#include <sys/ioctl.h>
72cf2d4f 37#include <sys/queue.h>
c5e97233 38#ifndef __DragonFly__
7674e7bf
FB		 39#include <sys/disk.h>
40#endif
c5e97233 41#endif
7674e7bf 42
49dc768d
AL		 43#ifdef _WIN32
44#include <windows.h>
45#endif
46
7d4b4ba5 47static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
f141eafe
AL		 48static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
49 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
c87c0672 50 BlockDriverCompletionFunc *cb, void *opaque);
f141eafe
AL		 51static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
52 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc 53 BlockDriverCompletionFunc *cb, void *opaque);
b2e12bc6
CH		 54static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
55 BlockDriverCompletionFunc *cb, void *opaque);
016f5cf6
AG		 56static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
57 BlockDriverCompletionFunc *cb, void *opaque);
5fafdf24 58static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
83f64091
FB		 59 uint8_t *buf, int nb_sectors);
60static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
61 const uint8_t *buf, int nb_sectors);
68485420
KW		 62static BlockDriverAIOCB *bdrv_co_aio_readv_em(BlockDriverState *bs,
63 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
64 BlockDriverCompletionFunc *cb, void *opaque);
65static BlockDriverAIOCB *bdrv_co_aio_writev_em(BlockDriverState *bs,
66 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
67 BlockDriverCompletionFunc *cb, void *opaque);
f9f05dc5
KW		 68static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
69 int64_t sector_num, int nb_sectors,
70 QEMUIOVector *iov);
71static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
72 int64_t sector_num, int nb_sectors,
73 QEMUIOVector *iov);
e7a8a783 74static int coroutine_fn bdrv_co_flush_em(BlockDriverState *bs);
ec530c81 75
1b7bdbc1
SH		 76static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(bdrv_states);
7ee930d0 78
8a22f02a
SH		 79static QLIST_HEAD(, BlockDriver) bdrv_drivers =
80 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 81
f9092b10
MA		 82/* The device to use for VM snapshots */
83static BlockDriverState *bs_snapshots;
84
eb852011
MA		 85/* If non-zero, use only whitelisted block drivers */
86static int use_bdrv_whitelist;
87
9e0b22f4
SH		 88#ifdef _WIN32
89static int is_windows_drive_prefix(const char *filename)
90{
91 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
92 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
93 filename[1] == ':');
94}
95
96int is_windows_drive(const char *filename)
97{
98 if (is_windows_drive_prefix(filename) &&
99 filename[2] == '\0')
100 return 1;
101 if (strstart(filename, "\\\\.\\", NULL) ||
102 strstart(filename, "//./", NULL))
103 return 1;
104 return 0;
105}
106#endif
107
108/* check if the path starts with "<protocol>:" */
109static int path_has_protocol(const char *path)
110{
111#ifdef _WIN32
112 if (is_windows_drive(path) ||
113 is_windows_drive_prefix(path)) {
114 return 0;
115 }
116#endif
117
118 return strchr(path, ':') != NULL;
119}
120
83f64091 121int path_is_absolute(const char *path)
3b0d4f61 122{
83f64091 123 const char *p;
21664424
FB		 124#ifdef _WIN32
125 /* specific case for names like: "\\.\d:" */
126 if (*path == '/' || *path == '\\')
127 return 1;
128#endif
83f64091
FB		 129 p = strchr(path, ':');
130 if (p)
131 p++;
132 else
133 p = path;
3b9f94e1
FB		 134#ifdef _WIN32
135 return (*p == '/' || *p == '\\');
136#else
137 return (*p == '/');
138#endif
3b0d4f61
FB		 139}
140
83f64091
FB		 141/* if filename is absolute, just copy it to dest. Otherwise, build a
142 path to it by considering it is relative to base_path. URL are
143 supported. */
144void path_combine(char *dest, int dest_size,
145 const char *base_path,
146 const char *filename)
3b0d4f61 147{
83f64091
FB		 148 const char *p, *p1;
149 int len;
150
151 if (dest_size <= 0)
152 return;
153 if (path_is_absolute(filename)) {
154 pstrcpy(dest, dest_size, filename);
155 } else {
156 p = strchr(base_path, ':');
157 if (p)
158 p++;
159 else
160 p = base_path;
3b9f94e1
FB		 161 p1 = strrchr(base_path, '/');
162#ifdef _WIN32
163 {
164 const char *p2;
165 p2 = strrchr(base_path, '\\');
166 if (!p1 || p2 > p1)
167 p1 = p2;
168 }
169#endif
83f64091
FB		 170 if (p1)
171 p1++;
172 else
173 p1 = base_path;
174 if (p1 > p)
175 p = p1;
176 len = p - base_path;
177 if (len > dest_size - 1)
178 len = dest_size - 1;
179 memcpy(dest, base_path, len);
180 dest[len] = '\0';
181 pstrcat(dest, dest_size, filename);
3b0d4f61 182 }
3b0d4f61
FB		 183}
184
5efa9d5a 185void bdrv_register(BlockDriver *bdrv)
ea2384d3 186{
68485420
KW		 187 if (bdrv->bdrv_co_readv) {
188 /* Emulate AIO by coroutines, and sync by AIO */
189 bdrv->bdrv_aio_readv = bdrv_co_aio_readv_em;
190 bdrv->bdrv_aio_writev = bdrv_co_aio_writev_em;
191 bdrv->bdrv_read = bdrv_read_em;
192 bdrv->bdrv_write = bdrv_write_em;
f9f05dc5
KW		 193 } else {
194 bdrv->bdrv_co_readv = bdrv_co_readv_em;
195 bdrv->bdrv_co_writev = bdrv_co_writev_em;
196
197 if (!bdrv->bdrv_aio_readv) {
198 /* add AIO emulation layer */
199 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
200 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
201 } else if (!bdrv->bdrv_read) {
202 /* add synchronous IO emulation layer */
203 bdrv->bdrv_read = bdrv_read_em;
204 bdrv->bdrv_write = bdrv_write_em;
205 }
83f64091 206 }
b2e12bc6
CH		 207
208 if (!bdrv->bdrv_aio_flush)
209 bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
210
8a22f02a 211 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 212}
b338082b
FB		 213
214/* create a new block device (by default it is empty) */
215BlockDriverState *bdrv_new(const char *device_name)
216{
1b7bdbc1 217 BlockDriverState *bs;
b338082b 218
7267c094 219 bs = g_malloc0(sizeof(BlockDriverState));
b338082b 220 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
ea2384d3 221 if (device_name[0] != '\0') {
1b7bdbc1 222 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
ea2384d3 223 }
b338082b
FB		 224 return bs;
225}
226
ea2384d3
FB		 227BlockDriver *bdrv_find_format(const char *format_name)
228{
229 BlockDriver *drv1;
8a22f02a
SH		 230 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
231 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 232 return drv1;
8a22f02a 233 }
ea2384d3
FB		 234 }
235 return NULL;
236}
237
eb852011
MA		 238static int bdrv_is_whitelisted(BlockDriver *drv)
239{
240 static const char *whitelist[] = {
241 CONFIG_BDRV_WHITELIST
242 };
243 const char **p;
244
245 if (!whitelist[0])
246 return 1; /* no whitelist, anything goes */
247
248 for (p = whitelist; *p; p++) {
249 if (!strcmp(drv->format_name, *p)) {
250 return 1;
251 }
252 }
253 return 0;
254}
255
256BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
257{
258 BlockDriver *drv = bdrv_find_format(format_name);
259 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
260}
261
0e7e1989
KW		 262int bdrv_create(BlockDriver *drv, const char* filename,
263 QEMUOptionParameter *options)
ea2384d3
FB		 264{
265 if (!drv->bdrv_create)
266 return -ENOTSUP;
0e7e1989
KW		 267
268 return drv->bdrv_create(filename, options);
ea2384d3
FB		 269}
270
84a12e66
CH		 271int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
272{
273 BlockDriver *drv;
274
b50cbabc 275 drv = bdrv_find_protocol(filename);
84a12e66 276 if (drv == NULL) {
16905d71 277 return -ENOENT;
84a12e66
CH		 278 }
279
280 return bdrv_create(drv, filename, options);
281}
282
d5249393 283#ifdef _WIN32
95389c86 284void get_tmp_filename(char *filename, int size)
d5249393 285{
3b9f94e1 286 char temp_dir[MAX_PATH];
3b46e624 287
3b9f94e1
FB		 288 GetTempPath(MAX_PATH, temp_dir);
289 GetTempFileName(temp_dir, "qem", 0, filename);
d5249393
FB		 290}
291#else
95389c86 292void get_tmp_filename(char *filename, int size)
fc01f7e7 293{
67b915a5 294 int fd;
7ccfb2eb 295 const char *tmpdir;
d5249393 296 /* XXX: race condition possible */
0badc1ee
AJ		 297 tmpdir = getenv("TMPDIR");
298 if (!tmpdir)
299 tmpdir = "/tmp";
300 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
ea2384d3
FB		 301 fd = mkstemp(filename);
302 close(fd);
303}
d5249393 304#endif
fc01f7e7 305
84a12e66
CH		 306/*
307 * Detect host devices. By convention, /dev/cdrom[N] is always
308 * recognized as a host CDROM.
309 */
310static BlockDriver *find_hdev_driver(const char *filename)
311{
312 int score_max = 0, score;
313 BlockDriver *drv = NULL, *d;
314
315 QLIST_FOREACH(d, &bdrv_drivers, list) {
316 if (d->bdrv_probe_device) {
317 score = d->bdrv_probe_device(filename);
318 if (score > score_max) {
319 score_max = score;
320 drv = d;
321 }
322 }
323 }
324
325 return drv;
326}
327
b50cbabc 328BlockDriver *bdrv_find_protocol(const char *filename)
83f64091
FB		 329{
330 BlockDriver *drv1;
331 char protocol[128];
1cec71e3 332 int len;
83f64091 333 const char *p;
19cb3738 334
66f82cee
KW		 335 /* TODO Drivers without bdrv_file_open must be specified explicitly */
336
39508e7a
CH		 337 /*
338 * XXX(hch): we really should not let host device detection
339 * override an explicit protocol specification, but moving this
340 * later breaks access to device names with colons in them.
341 * Thanks to the brain-dead persistent naming schemes on udev-
342 * based Linux systems those actually are quite common.
343 */
344 drv1 = find_hdev_driver(filename);
345 if (drv1) {
346 return drv1;
347 }
348
9e0b22f4 349 if (!path_has_protocol(filename)) {
39508e7a 350 return bdrv_find_format("file");
84a12e66 351 }
9e0b22f4
SH		 352 p = strchr(filename, ':');
353 assert(p != NULL);
1cec71e3
AL		 354 len = p - filename;
355 if (len > sizeof(protocol) - 1)
356 len = sizeof(protocol) - 1;
357 memcpy(protocol, filename, len);
358 protocol[len] = '\0';
8a22f02a 359 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
5fafdf24 360 if (drv1->protocol_name &&
8a22f02a 361 !strcmp(drv1->protocol_name, protocol)) {
83f64091 362 return drv1;
8a22f02a 363 }
83f64091
FB		 364 }
365 return NULL;
366}
367
c98ac35d 368static int find_image_format(const char *filename, BlockDriver **pdrv)
f3a5d3f8
CH		 369{
370 int ret, score, score_max;
371 BlockDriver *drv1, *drv;
372 uint8_t buf[2048];
373 BlockDriverState *bs;
374
f5edb014 375 ret = bdrv_file_open(&bs, filename, 0);
c98ac35d
SW		 376 if (ret < 0) {
377 *pdrv = NULL;
378 return ret;
379 }
f8ea0b00 380
08a00559
KW		 381 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
382 if (bs->sg || !bdrv_is_inserted(bs)) {
1a396859 383 bdrv_delete(bs);
c98ac35d
SW		 384 drv = bdrv_find_format("raw");
385 if (!drv) {
386 ret = -ENOENT;
387 }
388 *pdrv = drv;
389 return ret;
1a396859 390 }
f8ea0b00 391
83f64091
FB		 392 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
393 bdrv_delete(bs);
394 if (ret < 0) {
c98ac35d
SW		 395 *pdrv = NULL;
396 return ret;
83f64091
FB		 397 }
398
ea2384d3 399 score_max = 0;
84a12e66 400 drv = NULL;
8a22f02a 401 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
83f64091
FB		 402 if (drv1->bdrv_probe) {
403 score = drv1->bdrv_probe(buf, ret, filename);
404 if (score > score_max) {
405 score_max = score;
406 drv = drv1;
407 }
0849bf08 408 }
fc01f7e7 409 }
c98ac35d
SW		 410 if (!drv) {
411 ret = -ENOENT;
412 }
413 *pdrv = drv;
414 return ret;
ea2384d3
FB		 415}
416
51762288
SH		 417/**
418 * Set the current 'total_sectors' value
419 */
420static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
421{
422 BlockDriver *drv = bs->drv;
423
396759ad
NB		 424 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
425 if (bs->sg)
426 return 0;
427
51762288
SH		 428 /* query actual device if possible, otherwise just trust the hint */
429 if (drv->bdrv_getlength) {
430 int64_t length = drv->bdrv_getlength(bs);
431 if (length < 0) {
432 return length;
433 }
434 hint = length >> BDRV_SECTOR_BITS;
435 }
436
437 bs->total_sectors = hint;
438 return 0;
439}
440
c3993cdc
SH		 441/**
442 * Set open flags for a given cache mode
443 *
444 * Return 0 on success, -1 if the cache mode was invalid.
445 */
446int bdrv_parse_cache_flags(const char *mode, int *flags)
447{
448 *flags &= ~BDRV_O_CACHE_MASK;
449
450 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
451 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
92196b2f
SH		 452 } else if (!strcmp(mode, "directsync")) {
453 *flags |= BDRV_O_NOCACHE;
c3993cdc
SH		 454 } else if (!strcmp(mode, "writeback")) {
455 *flags |= BDRV_O_CACHE_WB;
456 } else if (!strcmp(mode, "unsafe")) {
457 *flags |= BDRV_O_CACHE_WB;
458 *flags |= BDRV_O_NO_FLUSH;
459 } else if (!strcmp(mode, "writethrough")) {
460 /* this is the default */
461 } else {
462 return -1;
463 }
464
465 return 0;
466}
467
57915332
KW		 468/*
469 * Common part for opening disk images and files
470 */
471static int bdrv_open_common(BlockDriverState *bs, const char *filename,
472 int flags, BlockDriver *drv)
473{
474 int ret, open_flags;
475
476 assert(drv != NULL);
477
66f82cee 478 bs->file = NULL;
51762288 479 bs->total_sectors = 0;
57915332
KW		 480 bs->encrypted = 0;
481 bs->valid_key = 0;
482 bs->open_flags = flags;
57915332
KW		 483 bs->buffer_alignment = 512;
484
485 pstrcpy(bs->filename, sizeof(bs->filename), filename);
486
487 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
488 return -ENOTSUP;
489 }
490
491 bs->drv = drv;
7267c094 492 bs->opaque = g_malloc0(drv->instance_size);
57915332 493
a6599793 494 if (flags & BDRV_O_CACHE_WB)
57915332
KW		 495 bs->enable_write_cache = 1;
496
497 /*
498 * Clear flags that are internal to the block layer before opening the
499 * image.
500 */
501 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
502
503 /*
ebabb67a 504 * Snapshots should be writable.
57915332
KW		 505 */
506 if (bs->is_temporary) {
507 open_flags |= BDRV_O_RDWR;
508 }
509
66f82cee
KW		 510 /* Open the image, either directly or using a protocol */
511 if (drv->bdrv_file_open) {
512 ret = drv->bdrv_file_open(bs, filename, open_flags);
513 } else {
514 ret = bdrv_file_open(&bs->file, filename, open_flags);
515 if (ret >= 0) {
516 ret = drv->bdrv_open(bs, open_flags);
517 }
518 }
519
57915332
KW		 520 if (ret < 0) {
521 goto free_and_fail;
522 }
523
524 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
51762288
SH		 525
526 ret = refresh_total_sectors(bs, bs->total_sectors);
527 if (ret < 0) {
528 goto free_and_fail;
57915332 529 }
51762288 530
57915332
KW		 531#ifndef _WIN32
532 if (bs->is_temporary) {
533 unlink(filename);
534 }
535#endif
536 return 0;
537
538free_and_fail:
66f82cee
KW		 539 if (bs->file) {
540 bdrv_delete(bs->file);
541 bs->file = NULL;
542 }
7267c094 543 g_free(bs->opaque);
57915332
KW		 544 bs->opaque = NULL;
545 bs->drv = NULL;
546 return ret;
547}
548
b6ce07aa
KW		 549/*
550 * Opens a file using a protocol (file, host_device, nbd, ...)
551 */
83f64091 552int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
ea2384d3 553{
83f64091 554 BlockDriverState *bs;
6db95603 555 BlockDriver *drv;
83f64091
FB		 556 int ret;
557
b50cbabc 558 drv = bdrv_find_protocol(filename);
6db95603
CH		 559 if (!drv) {
560 return -ENOENT;
561 }
562
83f64091 563 bs = bdrv_new("");
b6ce07aa 564 ret = bdrv_open_common(bs, filename, flags, drv);
83f64091
FB		 565 if (ret < 0) {
566 bdrv_delete(bs);
567 return ret;
3b0d4f61 568 }
71d0770c 569 bs->growable = 1;
83f64091
FB		 570 *pbs = bs;
571 return 0;
572}
573
b6ce07aa
KW		 574/*
575 * Opens a disk image (raw, qcow2, vmdk, ...)
576 */
d6e9098e
KW		 577int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
578 BlockDriver *drv)
ea2384d3 579{
b6ce07aa 580 int ret;
712e7874 581
83f64091 582 if (flags & BDRV_O_SNAPSHOT) {
ea2384d3
FB		 583 BlockDriverState *bs1;
584 int64_t total_size;
7c96d46e 585 int is_protocol = 0;
91a073a9
KW		 586 BlockDriver *bdrv_qcow2;
587 QEMUOptionParameter *options;
b6ce07aa
KW		 588 char tmp_filename[PATH_MAX];
589 char backing_filename[PATH_MAX];
3b46e624 590
ea2384d3
FB		 591 /* if snapshot, we create a temporary backing file and open it
592 instead of opening 'filename' directly */
33e3963e 593
ea2384d3
FB		 594 /* if there is a backing file, use it */
595 bs1 = bdrv_new("");
d6e9098e 596 ret = bdrv_open(bs1, filename, 0, drv);
51d7c00c 597 if (ret < 0) {
ea2384d3 598 bdrv_delete(bs1);
51d7c00c 599 return ret;
ea2384d3 600 }
3e82990b 601 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
7c96d46e
AL		 602
603 if (bs1->drv && bs1->drv->protocol_name)
604 is_protocol = 1;
605
ea2384d3 606 bdrv_delete(bs1);
3b46e624 607
ea2384d3 608 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
7c96d46e
AL		 609
610 /* Real path is meaningless for protocols */
611 if (is_protocol)
612 snprintf(backing_filename, sizeof(backing_filename),
613 "%s", filename);
114cdfa9
KS		 614 else if (!realpath(filename, backing_filename))
615 return -errno;
7c96d46e 616
91a073a9
KW		 617 bdrv_qcow2 = bdrv_find_format("qcow2");
618 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
619
3e82990b 620 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
91a073a9
KW		 621 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
622 if (drv) {
623 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
624 drv->format_name);
625 }
626
627 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
d748768c 628 free_option_parameters(options);
51d7c00c
AL		 629 if (ret < 0) {
630 return ret;
ea2384d3 631 }
91a073a9 632
ea2384d3 633 filename = tmp_filename;
91a073a9 634 drv = bdrv_qcow2;
ea2384d3
FB		 635 bs->is_temporary = 1;
636 }
712e7874 637
b6ce07aa 638 /* Find the right image format driver */
6db95603 639 if (!drv) {
c98ac35d 640 ret = find_image_format(filename, &drv);
51d7c00c 641 }
6987307c 642
51d7c00c 643 if (!drv) {
51d7c00c 644 goto unlink_and_fail;
ea2384d3 645 }
b6ce07aa
KW		 646
647 /* Open the image */
648 ret = bdrv_open_common(bs, filename, flags, drv);
649 if (ret < 0) {
6987307c
CH		 650 goto unlink_and_fail;
651 }
652
b6ce07aa
KW		 653 /* If there is a backing file, use it */
654 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
655 char backing_filename[PATH_MAX];
656 int back_flags;
657 BlockDriver *back_drv = NULL;
658
659 bs->backing_hd = bdrv_new("");
df2dbb4a
SH		 660
661 if (path_has_protocol(bs->backing_file)) {
662 pstrcpy(backing_filename, sizeof(backing_filename),
663 bs->backing_file);
664 } else {
665 path_combine(backing_filename, sizeof(backing_filename),
666 filename, bs->backing_file);
667 }
668
669 if (bs->backing_format[0] != '\0') {
b6ce07aa 670 back_drv = bdrv_find_format(bs->backing_format);
df2dbb4a 671 }
b6ce07aa
KW		 672
673 /* backing files always opened read-only */
674 back_flags =
675 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
676
677 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
678 if (ret < 0) {
679 bdrv_close(bs);
680 return ret;
681 }
682 if (bs->is_temporary) {
683 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
684 } else {
685 /* base image inherits from "parent" */
686 bs->backing_hd->keep_read_only = bs->keep_read_only;
687 }
688 }
689
690 if (!bdrv_key_required(bs)) {
7d4b4ba5 691 bdrv_dev_change_media_cb(bs, true);
b6ce07aa
KW		 692 }
693
694 return 0;
695
696unlink_and_fail:
697 if (bs->is_temporary) {
698 unlink(filename);
699 }
700 return ret;
701}
702
fc01f7e7
FB		 703void bdrv_close(BlockDriverState *bs)
704{
19cb3738 705 if (bs->drv) {
f9092b10
MA		 706 if (bs == bs_snapshots) {
707 bs_snapshots = NULL;
708 }
557df6ac 709 if (bs->backing_hd) {
ea2384d3 710 bdrv_delete(bs->backing_hd);
557df6ac
SH		 711 bs->backing_hd = NULL;
712 }
ea2384d3 713 bs->drv->bdrv_close(bs);
7267c094 714 g_free(bs->opaque);
ea2384d3
FB		 715#ifdef _WIN32
716 if (bs->is_temporary) {
717 unlink(bs->filename);
718 }
67b915a5 719#endif
ea2384d3
FB		 720 bs->opaque = NULL;
721 bs->drv = NULL;
b338082b 722
66f82cee
KW		 723 if (bs->file != NULL) {
724 bdrv_close(bs->file);
725 }
726
7d4b4ba5 727 bdrv_dev_change_media_cb(bs, false);
b338082b
FB		 728 }
729}
730
2bc93fed
MK		 731void bdrv_close_all(void)
732{
733 BlockDriverState *bs;
734
735 QTAILQ_FOREACH(bs, &bdrv_states, list) {
736 bdrv_close(bs);
737 }
738}
739
d22b2f41
RH		 740/* make a BlockDriverState anonymous by removing from bdrv_state list.
741 Also, NULL terminate the device_name to prevent double remove */
742void bdrv_make_anon(BlockDriverState *bs)
743{
744 if (bs->device_name[0] != '\0') {
745 QTAILQ_REMOVE(&bdrv_states, bs, list);
746 }
747 bs->device_name[0] = '\0';
748}
749
b338082b
FB		 750void bdrv_delete(BlockDriverState *bs)
751{
fa879d62 752 assert(!bs->dev);
18846dee 753
1b7bdbc1 754 /* remove from list, if necessary */
d22b2f41 755 bdrv_make_anon(bs);
34c6f050 756
b338082b 757 bdrv_close(bs);
66f82cee
KW		 758 if (bs->file != NULL) {
759 bdrv_delete(bs->file);
760 }
761
f9092b10 762 assert(bs != bs_snapshots);
7267c094 763 g_free(bs);
fc01f7e7
FB		 764}
765
fa879d62
MA		 766int bdrv_attach_dev(BlockDriverState *bs, void *dev)
767/* TODO change to DeviceState *dev when all users are qdevified */
18846dee 768{
fa879d62 769 if (bs->dev) {
18846dee
MA		 770 return -EBUSY;
771 }
fa879d62 772 bs->dev = dev;
18846dee
MA		 773 return 0;
774}
775
fa879d62
MA		 776/* TODO qdevified devices don't use this, remove when devices are qdevified */
777void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
18846dee 778{
fa879d62
MA		 779 if (bdrv_attach_dev(bs, dev) < 0) {
780 abort();
781 }
782}
783
784void bdrv_detach_dev(BlockDriverState *bs, void *dev)
785/* TODO change to DeviceState *dev when all users are qdevified */
786{
787 assert(bs->dev == dev);
788 bs->dev = NULL;
0e49de52
MA		 789 bs->dev_ops = NULL;
790 bs->dev_opaque = NULL;
29e05f20 791 bs->buffer_alignment = 512;
18846dee
MA		 792}
793
fa879d62
MA		 794/* TODO change to return DeviceState * when all users are qdevified */
795void *bdrv_get_attached_dev(BlockDriverState *bs)
18846dee 796{
fa879d62 797 return bs->dev;
18846dee
MA		 798}
799
0e49de52
MA		 800void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
801 void *opaque)
802{
803 bs->dev_ops = ops;
804 bs->dev_opaque = opaque;
2c6942fa
MA		 805 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
806 bs_snapshots = NULL;
807 }
0e49de52
MA		 808}
809
7d4b4ba5 810static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
0e49de52 811{
145feb17 812 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
7d4b4ba5 813 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
145feb17
MA		 814 }
815}
816
2c6942fa
MA		 817bool bdrv_dev_has_removable_media(BlockDriverState *bs)
818{
819 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
820}
821
e4def80b
MA		 822bool bdrv_dev_is_tray_open(BlockDriverState *bs)
823{
824 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
825 return bs->dev_ops->is_tray_open(bs->dev_opaque);
826 }
827 return false;
828}
829
145feb17
MA		 830static void bdrv_dev_resize_cb(BlockDriverState *bs)
831{
832 if (bs->dev_ops && bs->dev_ops->resize_cb) {
833 bs->dev_ops->resize_cb(bs->dev_opaque);
0e49de52
MA		 834 }
835}
836
f107639a
MA		 837bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
838{
839 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
840 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
841 }
842 return false;
843}
844
e97fc193
AL		 845/*
846 * Run consistency checks on an image
847 *
e076f338 848 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 849 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 850 * check are stored in res.
e97fc193 851 */
e076f338 852int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
e97fc193
AL		 853{
854 if (bs->drv->bdrv_check == NULL) {
855 return -ENOTSUP;
856 }
857
e076f338 858 memset(res, 0, sizeof(*res));
9ac228e0 859 return bs->drv->bdrv_check(bs, res);
e97fc193
AL		 860}
861
8a426614
KW		 862#define COMMIT_BUF_SECTORS 2048
863
33e3963e
FB		 864/* commit COW file into the raw image */
865int bdrv_commit(BlockDriverState *bs)
866{
19cb3738 867 BlockDriver *drv = bs->drv;
ee181196 868 BlockDriver *backing_drv;
8a426614
KW		 869 int64_t sector, total_sectors;
870 int n, ro, open_flags;
4dca4b63 871 int ret = 0, rw_ret = 0;
8a426614 872 uint8_t *buf;
4dca4b63
NS		 873 char filename[1024];
874 BlockDriverState *bs_rw, *bs_ro;
33e3963e 875
19cb3738
FB		 876 if (!drv)
877 return -ENOMEDIUM;
4dca4b63
NS		 878
879 if (!bs->backing_hd) {
880 return -ENOTSUP;
33e3963e
FB		 881 }
882
4dca4b63
NS		 883 if (bs->backing_hd->keep_read_only) {
884 return -EACCES;
885 }
ee181196
KW		 886
887 backing_drv = bs->backing_hd->drv;
4dca4b63
NS		 888 ro = bs->backing_hd->read_only;
889 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
890 open_flags = bs->backing_hd->open_flags;
891
892 if (ro) {
893 /* re-open as RW */
894 bdrv_delete(bs->backing_hd);
895 bs->backing_hd = NULL;
896 bs_rw = bdrv_new("");
ee181196
KW		 897 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
898 backing_drv);
4dca4b63
NS		 899 if (rw_ret < 0) {
900 bdrv_delete(bs_rw);
901 /* try to re-open read-only */
902 bs_ro = bdrv_new("");
ee181196
KW		 903 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
904 backing_drv);
4dca4b63
NS		 905 if (ret < 0) {
906 bdrv_delete(bs_ro);
907 /* drive not functional anymore */
908 bs->drv = NULL;
909 return ret;
910 }
911 bs->backing_hd = bs_ro;
912 return rw_ret;
913 }
914 bs->backing_hd = bs_rw;
ea2384d3 915 }
33e3963e 916
6ea44308 917 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
7267c094 918 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
8a426614
KW		 919
920 for (sector = 0; sector < total_sectors; sector += n) {
921 if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
922
923 if (bdrv_read(bs, sector, buf, n) != 0) {
924 ret = -EIO;
925 goto ro_cleanup;
926 }
927
928 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
929 ret = -EIO;
930 goto ro_cleanup;
931 }
ea2384d3 932 }
33e3963e 933 }
95389c86 934
1d44952f
CH		 935 if (drv->bdrv_make_empty) {
936 ret = drv->bdrv_make_empty(bs);
937 bdrv_flush(bs);
938 }
95389c86 939
3f5075ae
CH		 940 /*
941 * Make sure all data we wrote to the backing device is actually
942 * stable on disk.
943 */
944 if (bs->backing_hd)
945 bdrv_flush(bs->backing_hd);
4dca4b63
NS		 946
947ro_cleanup:
7267c094 948 g_free(buf);
4dca4b63
NS		 949
950 if (ro) {
951 /* re-open as RO */
952 bdrv_delete(bs->backing_hd);
953 bs->backing_hd = NULL;
954 bs_ro = bdrv_new("");
ee181196
KW		 955 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
956 backing_drv);
4dca4b63
NS		 957 if (ret < 0) {
958 bdrv_delete(bs_ro);
959 /* drive not functional anymore */
960 bs->drv = NULL;
961 return ret;
962 }
963 bs->backing_hd = bs_ro;
964 bs->backing_hd->keep_read_only = 0;
965 }
966
1d44952f 967 return ret;
33e3963e
FB		 968}
969
6ab4b5ab
MA		 970void bdrv_commit_all(void)
971{
972 BlockDriverState *bs;
973
974 QTAILQ_FOREACH(bs, &bdrv_states, list) {
975 bdrv_commit(bs);
976 }
977}
978
756e6736
KW		 979/*
980 * Return values:
981 * 0 - success
982 * -EINVAL - backing format specified, but no file
983 * -ENOSPC - can't update the backing file because no space is left in the
984 * image file header
985 * -ENOTSUP - format driver doesn't support changing the backing file
986 */
987int bdrv_change_backing_file(BlockDriverState *bs,
988 const char *backing_file, const char *backing_fmt)
989{
990 BlockDriver *drv = bs->drv;
991
992 if (drv->bdrv_change_backing_file != NULL) {
993 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
994 } else {
995 return -ENOTSUP;
996 }
997}
998
71d0770c
AL		 999static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1000 size_t size)
1001{
1002 int64_t len;
1003
1004 if (!bdrv_is_inserted(bs))
1005 return -ENOMEDIUM;
1006
1007 if (bs->growable)
1008 return 0;
1009
1010 len = bdrv_getlength(bs);
1011
fbb7b4e0
KW		 1012 if (offset < 0)
1013 return -EIO;
1014
1015 if ((offset > len) || (len - offset < size))
71d0770c
AL		 1016 return -EIO;
1017
1018 return 0;
1019}
1020
1021static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1022 int nb_sectors)
1023{
eb5a3165
JS		 1024 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1025 nb_sectors * BDRV_SECTOR_SIZE);
71d0770c
AL		 1026}
1027
e7a8a783
KW		 1028static inline bool bdrv_has_async_rw(BlockDriver *drv)
1029{
1030 return drv->bdrv_co_readv != bdrv_co_readv_em
1031 || drv->bdrv_aio_readv != bdrv_aio_readv_em;
1032}
1033
1034static inline bool bdrv_has_async_flush(BlockDriver *drv)
1035{
1036 return drv->bdrv_aio_flush != bdrv_aio_flush_em;
1037}
1038
19cb3738 1039/* return < 0 if error. See bdrv_write() for the return codes */
5fafdf24 1040int bdrv_read(BlockDriverState *bs, int64_t sector_num,
fc01f7e7
FB		 1041 uint8_t *buf, int nb_sectors)
1042{
ea2384d3
FB		 1043 BlockDriver *drv = bs->drv;
1044
19cb3738
FB		 1045 if (!drv)
1046 return -ENOMEDIUM;
e7a8a783
KW		 1047
1048 if (bdrv_has_async_rw(drv) && qemu_in_coroutine()) {
1049 QEMUIOVector qiov;
1050 struct iovec iov = {
1051 .iov_base = (void *)buf,
1052 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1053 };
1054
1055 qemu_iovec_init_external(&qiov, &iov, 1);
1056 return bdrv_co_readv(bs, sector_num, nb_sectors, &qiov);
1057 }
1058
71d0770c
AL		 1059 if (bdrv_check_request(bs, sector_num, nb_sectors))
1060 return -EIO;
b338082b 1061
eda578e5 1062 return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
fc01f7e7
FB		 1063}
1064
7cd1e32a 1065static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
a55eb92c 1066 int nb_sectors, int dirty)
7cd1e32a
LS		 1067{
1068 int64_t start, end;
c6d22830 1069 unsigned long val, idx, bit;
a55eb92c 1070
6ea44308 1071 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
c6d22830 1072 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
a55eb92c
JK		 1073
1074 for (; start <= end; start++) {
c6d22830
JK		 1075 idx = start / (sizeof(unsigned long) * 8);
1076 bit = start % (sizeof(unsigned long) * 8);
1077 val = bs->dirty_bitmap[idx];
1078 if (dirty) {
6d59fec1 1079 if (!(val & (1UL << bit))) {
aaa0eb75 1080 bs->dirty_count++;
6d59fec1 1081 val |= 1UL << bit;
aaa0eb75 1082 }
c6d22830 1083 } else {
6d59fec1 1084 if (val & (1UL << bit)) {
aaa0eb75 1085 bs->dirty_count--;
6d59fec1 1086 val &= ~(1UL << bit);
aaa0eb75 1087 }
c6d22830
JK		 1088 }
1089 bs->dirty_bitmap[idx] = val;
7cd1e32a
LS		 1090 }
1091}
1092
5fafdf24 1093/* Return < 0 if error. Important errors are:
19cb3738
FB		 1094 -EIO generic I/O error (may happen for all errors)
1095 -ENOMEDIUM No media inserted.
1096 -EINVAL Invalid sector number or nb_sectors
1097 -EACCES Trying to write a read-only device
1098*/
5fafdf24 1099int bdrv_write(BlockDriverState *bs, int64_t sector_num,
fc01f7e7
FB		 1100 const uint8_t *buf, int nb_sectors)
1101{
83f64091 1102 BlockDriver *drv = bs->drv;
e7a8a783 1103
19cb3738
FB		 1104 if (!bs->drv)
1105 return -ENOMEDIUM;
e7a8a783
KW		 1106
1107 if (bdrv_has_async_rw(drv) && qemu_in_coroutine()) {
1108 QEMUIOVector qiov;
1109 struct iovec iov = {
1110 .iov_base = (void *)buf,
1111 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1112 };
1113
1114 qemu_iovec_init_external(&qiov, &iov, 1);
1115 return bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1116 }
1117
0849bf08 1118 if (bs->read_only)
19cb3738 1119 return -EACCES;
71d0770c
AL		 1120 if (bdrv_check_request(bs, sector_num, nb_sectors))
1121 return -EIO;
a55eb92c 1122
c6d22830 1123 if (bs->dirty_bitmap) {
7cd1e32a
LS		 1124 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1125 }
a55eb92c 1126
294cc35f
KW		 1127 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1128 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1129 }
1130
42fb2807 1131 return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
83f64091
FB		 1132}
1133
eda578e5
AL		 1134int bdrv_pread(BlockDriverState *bs, int64_t offset,
1135 void *buf, int count1)
83f64091 1136{
6ea44308 1137 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
83f64091
FB		 1138 int len, nb_sectors, count;
1139 int64_t sector_num;
9a8c4cce 1140 int ret;
83f64091
FB		 1141
1142 count = count1;
1143 /* first read to align to sector start */
6ea44308 1144 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
83f64091
FB		 1145 if (len > count)
1146 len = count;
6ea44308 1147 sector_num = offset >> BDRV_SECTOR_BITS;
83f64091 1148 if (len > 0) {
9a8c4cce
KW		 1149 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1150 return ret;
6ea44308 1151 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
83f64091
FB		 1152 count -= len;
1153 if (count == 0)
1154 return count1;
1155 sector_num++;
1156 buf += len;
1157 }
1158
1159 /* read the sectors "in place" */
6ea44308 1160 nb_sectors = count >> BDRV_SECTOR_BITS;
83f64091 1161 if (nb_sectors > 0) {
9a8c4cce
KW		 1162 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1163 return ret;
83f64091 1164 sector_num += nb_sectors;
6ea44308 1165 len = nb_sectors << BDRV_SECTOR_BITS;
83f64091
FB		 1166 buf += len;
1167 count -= len;
1168 }
1169
1170 /* add data from the last sector */
1171 if (count > 0) {
9a8c4cce
KW		 1172 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1173 return ret;
83f64091
FB		 1174 memcpy(buf, tmp_buf, count);
1175 }
1176 return count1;
1177}
1178
eda578e5
AL		 1179int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1180 const void *buf, int count1)
83f64091 1181{
6ea44308 1182 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
83f64091
FB		 1183 int len, nb_sectors, count;
1184 int64_t sector_num;
9a8c4cce 1185 int ret;
83f64091
FB		 1186
1187 count = count1;
1188 /* first write to align to sector start */
6ea44308 1189 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
83f64091
FB		 1190 if (len > count)
1191 len = count;
6ea44308 1192 sector_num = offset >> BDRV_SECTOR_BITS;
83f64091 1193 if (len > 0) {
9a8c4cce
KW		 1194 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1195 return ret;
6ea44308 1196 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
9a8c4cce
KW		 1197 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1198 return ret;
83f64091
FB		 1199 count -= len;
1200 if (count == 0)
1201 return count1;
1202 sector_num++;
1203 buf += len;
1204 }
1205
1206 /* write the sectors "in place" */
6ea44308 1207 nb_sectors = count >> BDRV_SECTOR_BITS;
83f64091 1208 if (nb_sectors > 0) {
9a8c4cce
KW		 1209 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1210 return ret;
83f64091 1211 sector_num += nb_sectors;
6ea44308 1212 len = nb_sectors << BDRV_SECTOR_BITS;
83f64091
FB		 1213 buf += len;
1214 count -= len;
1215 }
1216
1217 /* add data from the last sector */
1218 if (count > 0) {
9a8c4cce
KW		 1219 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1220 return ret;
83f64091 1221 memcpy(tmp_buf, buf, count);
9a8c4cce
KW		 1222 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1223 return ret;
83f64091
FB		 1224 }
1225 return count1;
1226}
83f64091 1227
f08145fe
KW		 1228/*
1229 * Writes to the file and ensures that no writes are reordered across this
1230 * request (acts as a barrier)
1231 *
1232 * Returns 0 on success, -errno in error cases.
1233 */
1234int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1235 const void *buf, int count)
1236{
1237 int ret;
1238
1239 ret = bdrv_pwrite(bs, offset, buf, count);
1240 if (ret < 0) {
1241 return ret;
1242 }
1243
92196b2f
SH		 1244 /* No flush needed for cache modes that use O_DSYNC */
1245 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
f08145fe
KW		 1246 bdrv_flush(bs);
1247 }
1248
1249 return 0;
1250}
1251
da1fa91d
KW		 1252int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1253 int nb_sectors, QEMUIOVector *qiov)
1254{
1255 BlockDriver *drv = bs->drv;
1256
1257 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1258
1259 if (!drv) {
1260 return -ENOMEDIUM;
1261 }
1262 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1263 return -EIO;
1264 }
1265
1266 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1267}
1268
1269int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1270 int nb_sectors, QEMUIOVector *qiov)
1271{
1272 BlockDriver *drv = bs->drv;
1273
1274 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1275
1276 if (!bs->drv) {
1277 return -ENOMEDIUM;
1278 }
1279 if (bs->read_only) {
1280 return -EACCES;
1281 }
1282 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1283 return -EIO;
1284 }
1285
1286 if (bs->dirty_bitmap) {
1287 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1288 }
1289
1290 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1291 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1292 }
1293
1294 return drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1295}
1296
83f64091
FB		 1297/**
1298 * Truncate file to 'offset' bytes (needed only for file protocols)
1299 */
1300int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1301{
1302 BlockDriver *drv = bs->drv;
51762288 1303 int ret;
83f64091 1304 if (!drv)
19cb3738 1305 return -ENOMEDIUM;
83f64091
FB		 1306 if (!drv->bdrv_truncate)
1307 return -ENOTSUP;
59f2689d
NS		 1308 if (bs->read_only)
1309 return -EACCES;
8591675f
MT		 1310 if (bdrv_in_use(bs))
1311 return -EBUSY;
51762288
SH		 1312 ret = drv->bdrv_truncate(bs, offset);
1313 if (ret == 0) {
1314 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
145feb17 1315 bdrv_dev_resize_cb(bs);
51762288
SH		 1316 }
1317 return ret;
83f64091
FB		 1318}
1319
4a1d5e1f
FZ		 1320/**
1321 * Length of a allocated file in bytes. Sparse files are counted by actual
1322 * allocated space. Return < 0 if error or unknown.
1323 */
1324int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1325{
1326 BlockDriver *drv = bs->drv;
1327 if (!drv) {
1328 return -ENOMEDIUM;
1329 }
1330 if (drv->bdrv_get_allocated_file_size) {
1331 return drv->bdrv_get_allocated_file_size(bs);
1332 }
1333 if (bs->file) {
1334 return bdrv_get_allocated_file_size(bs->file);
1335 }
1336 return -ENOTSUP;
1337}
1338
83f64091
FB		 1339/**
1340 * Length of a file in bytes. Return < 0 if error or unknown.
1341 */
1342int64_t bdrv_getlength(BlockDriverState *bs)
1343{
1344 BlockDriver *drv = bs->drv;
1345 if (!drv)
19cb3738 1346 return -ENOMEDIUM;
51762288 1347
2c6942fa 1348 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
46a4e4e6
SH		 1349 if (drv->bdrv_getlength) {
1350 return drv->bdrv_getlength(bs);
1351 }
83f64091 1352 }
46a4e4e6 1353 return bs->total_sectors * BDRV_SECTOR_SIZE;
fc01f7e7
FB		 1354}
1355
19cb3738 1356/* return 0 as number of sectors if no device present or error */
96b8f136 1357void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
fc01f7e7 1358{
19cb3738
FB		 1359 int64_t length;
1360 length = bdrv_getlength(bs);
1361 if (length < 0)
1362 length = 0;
1363 else
6ea44308 1364 length = length >> BDRV_SECTOR_BITS;
19cb3738 1365 *nb_sectors_ptr = length;
fc01f7e7 1366}
cf98951b 1367
f3d54fc4
AL		 1368struct partition {
1369 uint8_t boot_ind; /* 0x80 - active */
1370 uint8_t head; /* starting head */
1371 uint8_t sector; /* starting sector */
1372 uint8_t cyl; /* starting cylinder */
1373 uint8_t sys_ind; /* What partition type */
1374 uint8_t end_head; /* end head */
1375 uint8_t end_sector; /* end sector */
1376 uint8_t end_cyl; /* end cylinder */
1377 uint32_t start_sect; /* starting sector counting from 0 */
1378 uint32_t nr_sects; /* nr of sectors in partition */
541dc0d4 1379} QEMU_PACKED;
f3d54fc4
AL		 1380
1381/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1382static int guess_disk_lchs(BlockDriverState *bs,
1383 int *pcylinders, int *pheads, int *psectors)
1384{
eb5a3165 1385 uint8_t buf[BDRV_SECTOR_SIZE];
f3d54fc4
AL		 1386 int ret, i, heads, sectors, cylinders;
1387 struct partition *p;
1388 uint32_t nr_sects;
a38131b6 1389 uint64_t nb_sectors;
f3d54fc4
AL		 1390
1391 bdrv_get_geometry(bs, &nb_sectors);
1392
1393 ret = bdrv_read(bs, 0, buf, 1);
1394 if (ret < 0)
1395 return -1;
1396 /* test msdos magic */
1397 if (buf[510] != 0x55 || buf[511] != 0xaa)
1398 return -1;
1399 for(i = 0; i < 4; i++) {
1400 p = ((struct partition *)(buf + 0x1be)) + i;
1401 nr_sects = le32_to_cpu(p->nr_sects);
1402 if (nr_sects && p->end_head) {
1403 /* We make the assumption that the partition terminates on
1404 a cylinder boundary */
1405 heads = p->end_head + 1;
1406 sectors = p->end_sector & 63;
1407 if (sectors == 0)
1408 continue;
1409 cylinders = nb_sectors / (heads * sectors);
1410 if (cylinders < 1 || cylinders > 16383)
1411 continue;
1412 *pheads = heads;
1413 *psectors = sectors;
1414 *pcylinders = cylinders;
1415#if 0
1416 printf("guessed geometry: LCHS=%d %d %d\n",
1417 cylinders, heads, sectors);
1418#endif
1419 return 0;
1420 }
1421 }
1422 return -1;
1423}
1424
1425void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1426{
1427 int translation, lba_detected = 0;
1428 int cylinders, heads, secs;
a38131b6 1429 uint64_t nb_sectors;
f3d54fc4
AL		 1430
1431 /* if a geometry hint is available, use it */
1432 bdrv_get_geometry(bs, &nb_sectors);
1433 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1434 translation = bdrv_get_translation_hint(bs);
1435 if (cylinders != 0) {
1436 *pcyls = cylinders;
1437 *pheads = heads;
1438 *psecs = secs;
1439 } else {
1440 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1441 if (heads > 16) {
1442 /* if heads > 16, it means that a BIOS LBA
1443 translation was active, so the default
1444 hardware geometry is OK */
1445 lba_detected = 1;
1446 goto default_geometry;
1447 } else {
1448 *pcyls = cylinders;
1449 *pheads = heads;
1450 *psecs = secs;
1451 /* disable any translation to be in sync with
1452 the logical geometry */
1453 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1454 bdrv_set_translation_hint(bs,
1455 BIOS_ATA_TRANSLATION_NONE);
1456 }
1457 }
1458 } else {
1459 default_geometry:
1460 /* if no geometry, use a standard physical disk geometry */
1461 cylinders = nb_sectors / (16 * 63);
1462
1463 if (cylinders > 16383)
1464 cylinders = 16383;
1465 else if (cylinders < 2)
1466 cylinders = 2;
1467 *pcyls = cylinders;
1468 *pheads = 16;
1469 *psecs = 63;
1470 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1471 if ((*pcyls * *pheads) <= 131072) {
1472 bdrv_set_translation_hint(bs,
1473 BIOS_ATA_TRANSLATION_LARGE);
1474 } else {
1475 bdrv_set_translation_hint(bs,
1476 BIOS_ATA_TRANSLATION_LBA);
1477 }
1478 }
1479 }
1480 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1481 }
1482}
1483
5fafdf24 1484void bdrv_set_geometry_hint(BlockDriverState *bs,
b338082b
FB		 1485 int cyls, int heads, int secs)
1486{
1487 bs->cyls = cyls;
1488 bs->heads = heads;
1489 bs->secs = secs;
1490}
1491
46d4767d
FB		 1492void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1493{
1494 bs->translation = translation;
1495}
1496
5fafdf24 1497void bdrv_get_geometry_hint(BlockDriverState *bs,
b338082b
FB		 1498 int *pcyls, int *pheads, int *psecs)
1499{
1500 *pcyls = bs->cyls;
1501 *pheads = bs->heads;
1502 *psecs = bs->secs;
1503}
1504
5bbdbb46
BS		 1505/* Recognize floppy formats */
1506typedef struct FDFormat {
1507 FDriveType drive;
1508 uint8_t last_sect;
1509 uint8_t max_track;
1510 uint8_t max_head;
1511} FDFormat;
1512
1513static const FDFormat fd_formats[] = {
1514 /* First entry is default format */
1515 /* 1.44 MB 3"1/2 floppy disks */
1516 { FDRIVE_DRV_144, 18, 80, 1, },
1517 { FDRIVE_DRV_144, 20, 80, 1, },
1518 { FDRIVE_DRV_144, 21, 80, 1, },
1519 { FDRIVE_DRV_144, 21, 82, 1, },
1520 { FDRIVE_DRV_144, 21, 83, 1, },
1521 { FDRIVE_DRV_144, 22, 80, 1, },
1522 { FDRIVE_DRV_144, 23, 80, 1, },
1523 { FDRIVE_DRV_144, 24, 80, 1, },
1524 /* 2.88 MB 3"1/2 floppy disks */
1525 { FDRIVE_DRV_288, 36, 80, 1, },
1526 { FDRIVE_DRV_288, 39, 80, 1, },
1527 { FDRIVE_DRV_288, 40, 80, 1, },
1528 { FDRIVE_DRV_288, 44, 80, 1, },
1529 { FDRIVE_DRV_288, 48, 80, 1, },
1530 /* 720 kB 3"1/2 floppy disks */
1531 { FDRIVE_DRV_144, 9, 80, 1, },
1532 { FDRIVE_DRV_144, 10, 80, 1, },
1533 { FDRIVE_DRV_144, 10, 82, 1, },
1534 { FDRIVE_DRV_144, 10, 83, 1, },
1535 { FDRIVE_DRV_144, 13, 80, 1, },
1536 { FDRIVE_DRV_144, 14, 80, 1, },
1537 /* 1.2 MB 5"1/4 floppy disks */
1538 { FDRIVE_DRV_120, 15, 80, 1, },
1539 { FDRIVE_DRV_120, 18, 80, 1, },
1540 { FDRIVE_DRV_120, 18, 82, 1, },
1541 { FDRIVE_DRV_120, 18, 83, 1, },
1542 { FDRIVE_DRV_120, 20, 80, 1, },
1543 /* 720 kB 5"1/4 floppy disks */
1544 { FDRIVE_DRV_120, 9, 80, 1, },
1545 { FDRIVE_DRV_120, 11, 80, 1, },
1546 /* 360 kB 5"1/4 floppy disks */
1547 { FDRIVE_DRV_120, 9, 40, 1, },
1548 { FDRIVE_DRV_120, 9, 40, 0, },
1549 { FDRIVE_DRV_120, 10, 41, 1, },
1550 { FDRIVE_DRV_120, 10, 42, 1, },
1551 /* 320 kB 5"1/4 floppy disks */
1552 { FDRIVE_DRV_120, 8, 40, 1, },
1553 { FDRIVE_DRV_120, 8, 40, 0, },
1554 /* 360 kB must match 5"1/4 better than 3"1/2... */
1555 { FDRIVE_DRV_144, 9, 80, 0, },
1556 /* end */
1557 { FDRIVE_DRV_NONE, -1, -1, 0, },
1558};
1559
1560void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
1561 int *max_track, int *last_sect,
1562 FDriveType drive_in, FDriveType *drive)
1563{
1564 const FDFormat *parse;
1565 uint64_t nb_sectors, size;
1566 int i, first_match, match;
1567
1568 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
1569 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
1570 /* User defined disk */
1571 } else {
1572 bdrv_get_geometry(bs, &nb_sectors);
1573 match = -1;
1574 first_match = -1;
1575 for (i = 0; ; i++) {
1576 parse = &fd_formats[i];
1577 if (parse->drive == FDRIVE_DRV_NONE) {
1578 break;
1579 }
1580 if (drive_in == parse->drive ||
1581 drive_in == FDRIVE_DRV_NONE) {
1582 size = (parse->max_head + 1) * parse->max_track *
1583 parse->last_sect;
1584 if (nb_sectors == size) {
1585 match = i;
1586 break;
1587 }
1588 if (first_match == -1) {
1589 first_match = i;
1590 }
1591 }
1592 }
1593 if (match == -1) {
1594 if (first_match == -1) {
1595 match = 1;
1596 } else {
1597 match = first_match;
1598 }
1599 parse = &fd_formats[match];
1600 }
1601 *nb_heads = parse->max_head + 1;
1602 *max_track = parse->max_track;
1603 *last_sect = parse->last_sect;
1604 *drive = parse->drive;
1605 }
1606}
1607
46d4767d
FB		 1608int bdrv_get_translation_hint(BlockDriverState *bs)
1609{
1610 return bs->translation;
1611}
1612
abd7f68d
MA		 1613void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1614 BlockErrorAction on_write_error)
1615{
1616 bs->on_read_error = on_read_error;
1617 bs->on_write_error = on_write_error;
1618}
1619
1620BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1621{
1622 return is_read ? bs->on_read_error : bs->on_write_error;
1623}
1624
b338082b
FB		 1625int bdrv_is_read_only(BlockDriverState *bs)
1626{
1627 return bs->read_only;
1628}
1629
985a03b0
TS		 1630int bdrv_is_sg(BlockDriverState *bs)
1631{
1632 return bs->sg;
1633}
1634
e900a7b7
CH		 1635int bdrv_enable_write_cache(BlockDriverState *bs)
1636{
1637 return bs->enable_write_cache;
1638}
1639
ea2384d3
FB		 1640int bdrv_is_encrypted(BlockDriverState *bs)
1641{
1642 if (bs->backing_hd && bs->backing_hd->encrypted)
1643 return 1;
1644 return bs->encrypted;
1645}
1646
c0f4ce77
AL		 1647int bdrv_key_required(BlockDriverState *bs)
1648{
1649 BlockDriverState *backing_hd = bs->backing_hd;
1650
1651 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1652 return 1;
1653 return (bs->encrypted && !bs->valid_key);
1654}
1655
ea2384d3
FB		 1656int bdrv_set_key(BlockDriverState *bs, const char *key)
1657{
1658 int ret;
1659 if (bs->backing_hd && bs->backing_hd->encrypted) {
1660 ret = bdrv_set_key(bs->backing_hd, key);
1661 if (ret < 0)
1662 return ret;
1663 if (!bs->encrypted)
1664 return 0;
1665 }
fd04a2ae
SH		 1666 if (!bs->encrypted) {
1667 return -EINVAL;
1668 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1669 return -ENOMEDIUM;
1670 }
c0f4ce77 1671 ret = bs->drv->bdrv_set_key(bs, key);
bb5fc20f
AL		 1672 if (ret < 0) {
1673 bs->valid_key = 0;
1674 } else if (!bs->valid_key) {
1675 bs->valid_key = 1;
1676 /* call the change callback now, we skipped it on open */
7d4b4ba5 1677 bdrv_dev_change_media_cb(bs, true);
bb5fc20f 1678 }
c0f4ce77 1679 return ret;
ea2384d3
FB		 1680}
1681
1682void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1683{
19cb3738 1684 if (!bs->drv) {
ea2384d3
FB		 1685 buf[0] = '\0';
1686 } else {
1687 pstrcpy(buf, buf_size, bs->drv->format_name);
1688 }
1689}
1690
5fafdf24 1691void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
ea2384d3
FB		 1692 void *opaque)
1693{
1694 BlockDriver *drv;
1695
8a22f02a 1696 QLIST_FOREACH(drv, &bdrv_drivers, list) {
ea2384d3
FB		 1697 it(opaque, drv->format_name);
1698 }
1699}
1700
b338082b
FB		 1701BlockDriverState *bdrv_find(const char *name)
1702{
1703 BlockDriverState *bs;
1704
1b7bdbc1
SH		 1705 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1706 if (!strcmp(name, bs->device_name)) {
b338082b 1707 return bs;
1b7bdbc1 1708 }
b338082b
FB		 1709 }
1710 return NULL;
1711}
1712
2f399b0a
MA		 1713BlockDriverState *bdrv_next(BlockDriverState *bs)
1714{
1715 if (!bs) {
1716 return QTAILQ_FIRST(&bdrv_states);
1717 }
1718 return QTAILQ_NEXT(bs, list);
1719}
1720
51de9760 1721void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
81d0912d
FB		 1722{
1723 BlockDriverState *bs;
1724
1b7bdbc1 1725 QTAILQ_FOREACH(bs, &bdrv_states, list) {
51de9760 1726 it(opaque, bs);
81d0912d
FB		 1727 }
1728}
1729
ea2384d3
FB		 1730const char *bdrv_get_device_name(BlockDriverState *bs)
1731{
1732 return bs->device_name;
1733}
1734
205ef796 1735int bdrv_flush(BlockDriverState *bs)
7a6cba61 1736{
016f5cf6 1737 if (bs->open_flags & BDRV_O_NO_FLUSH) {
205ef796
KW		 1738 return 0;
1739 }
1740
e7a8a783
KW		 1741 if (bs->drv && bdrv_has_async_flush(bs->drv) && qemu_in_coroutine()) {
1742 return bdrv_co_flush_em(bs);
1743 }
1744
205ef796
KW		 1745 if (bs->drv && bs->drv->bdrv_flush) {
1746 return bs->drv->bdrv_flush(bs);
016f5cf6
AG		 1747 }
1748
205ef796
KW		 1749 /*
1750 * Some block drivers always operate in either writethrough or unsafe mode
1751 * and don't support bdrv_flush therefore. Usually qemu doesn't know how
1752 * the server works (because the behaviour is hardcoded or depends on
1753 * server-side configuration), so we can't ensure that everything is safe
1754 * on disk. Returning an error doesn't work because that would break guests
1755 * even if the server operates in writethrough mode.
1756 *
1757 * Let's hope the user knows what he's doing.
1758 */
1759 return 0;
7a6cba61
PB		 1760}
1761
c6ca28d6
AL		 1762void bdrv_flush_all(void)
1763{
1764 BlockDriverState *bs;
1765
1b7bdbc1 1766 QTAILQ_FOREACH(bs, &bdrv_states, list) {
c602a489 1767 if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
c6ca28d6 1768 bdrv_flush(bs);
1b7bdbc1
SH		 1769 }
1770 }
c6ca28d6
AL		 1771}
1772
f2feebbd
KW		 1773int bdrv_has_zero_init(BlockDriverState *bs)
1774{
1775 assert(bs->drv);
1776
336c1c12
KW		 1777 if (bs->drv->bdrv_has_zero_init) {
1778 return bs->drv->bdrv_has_zero_init(bs);
f2feebbd
KW		 1779 }
1780
1781 return 1;
1782}
1783
bb8bf76f
CH		 1784int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
1785{
1786 if (!bs->drv) {
1787 return -ENOMEDIUM;
1788 }
1789 if (!bs->drv->bdrv_discard) {
1790 return 0;
1791 }
1792 return bs->drv->bdrv_discard(bs, sector_num, nb_sectors);
1793}
1794
f58c7b35
TS		 1795/*
1796 * Returns true iff the specified sector is present in the disk image. Drivers
1797 * not implementing the functionality are assumed to not support backing files,
1798 * hence all their sectors are reported as allocated.
1799 *
1800 * 'pnum' is set to the number of sectors (including and immediately following
1801 * the specified sector) that are known to be in the same
1802 * allocated/unallocated state.
1803 *
1804 * 'nb_sectors' is the max value 'pnum' should be set to.
1805 */
1806int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1807 int *pnum)
1808{
1809 int64_t n;
1810 if (!bs->drv->bdrv_is_allocated) {
1811 if (sector_num >= bs->total_sectors) {
1812 *pnum = 0;
1813 return 0;
1814 }
1815 n = bs->total_sectors - sector_num;
1816 *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1817 return 1;
1818 }
1819 return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1820}
1821
2582bfed
LC		 1822void bdrv_mon_event(const BlockDriverState *bdrv,
1823 BlockMonEventAction action, int is_read)
1824{
1825 QObject *data;
1826 const char *action_str;
1827
1828 switch (action) {
1829 case BDRV_ACTION_REPORT:
1830 action_str = "report";
1831 break;
1832 case BDRV_ACTION_IGNORE:
1833 action_str = "ignore";
1834 break;
1835 case BDRV_ACTION_STOP:
1836 action_str = "stop";
1837 break;
1838 default:
1839 abort();
1840 }
1841
1842 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1843 bdrv->device_name,
1844 action_str,
1845 is_read ? "read" : "write");
1846 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1847
1848 qobject_decref(data);
1849}
1850
d15e5465 1851static void bdrv_print_dict(QObject *obj, void *opaque)
b338082b 1852{
d15e5465
LC		 1853 QDict *bs_dict;
1854 Monitor *mon = opaque;
1855
1856 bs_dict = qobject_to_qdict(obj);
1857
d8aeeb31 1858 monitor_printf(mon, "%s: removable=%d",
d15e5465 1859 qdict_get_str(bs_dict, "device"),
d15e5465
LC		 1860 qdict_get_bool(bs_dict, "removable"));
1861
1862 if (qdict_get_bool(bs_dict, "removable")) {
1863 monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
e4def80b
MA		 1864 monitor_printf(mon, " tray-open=%d",
1865 qdict_get_bool(bs_dict, "tray-open"));
d15e5465 1866 }
d15e5465
LC		 1867 if (qdict_haskey(bs_dict, "inserted")) {
1868 QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1869
1870 monitor_printf(mon, " file=");
1871 monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1872 if (qdict_haskey(qdict, "backing_file")) {
1873 monitor_printf(mon, " backing_file=");
1874 monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1875 }
1876 monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1877 qdict_get_bool(qdict, "ro"),
1878 qdict_get_str(qdict, "drv"),
1879 qdict_get_bool(qdict, "encrypted"));
1880 } else {
1881 monitor_printf(mon, " [not inserted]");
1882 }
1883
1884 monitor_printf(mon, "\n");
1885}
1886
1887void bdrv_info_print(Monitor *mon, const QObject *data)
1888{
1889 qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1890}
1891
d15e5465
LC		 1892void bdrv_info(Monitor *mon, QObject **ret_data)
1893{
1894 QList *bs_list;
b338082b
FB		 1895 BlockDriverState *bs;
1896
d15e5465
LC		 1897 bs_list = qlist_new();
1898
1b7bdbc1 1899 QTAILQ_FOREACH(bs, &bdrv_states, list) {
d15e5465 1900 QObject *bs_obj;
e4def80b 1901 QDict *bs_dict;
d15e5465 1902
d8aeeb31 1903 bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': 'unknown', "
d15e5465 1904 "'removable': %i, 'locked': %i }",
2c6942fa
MA		 1905 bs->device_name,
1906 bdrv_dev_has_removable_media(bs),
f107639a 1907 bdrv_dev_is_medium_locked(bs));
e4def80b 1908 bs_dict = qobject_to_qdict(bs_obj);
d15e5465 1909
e4def80b
MA		 1910 if (bdrv_dev_has_removable_media(bs)) {
1911 qdict_put(bs_dict, "tray-open",
1912 qbool_from_int(bdrv_dev_is_tray_open(bs)));
1913 }
19cb3738 1914 if (bs->drv) {
d15e5465 1915 QObject *obj;
d15e5465
LC		 1916
1917 obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1918 "'encrypted': %i }",
1919 bs->filename, bs->read_only,
1920 bs->drv->format_name,
1921 bdrv_is_encrypted(bs));
fef30743 1922 if (bs->backing_file[0] != '\0') {
d15e5465
LC		 1923 QDict *qdict = qobject_to_qdict(obj);
1924 qdict_put(qdict, "backing_file",
1925 qstring_from_str(bs->backing_file));
376253ec 1926 }
d15e5465
LC		 1927
1928 qdict_put_obj(bs_dict, "inserted", obj);
b338082b 1929 }
d15e5465 1930 qlist_append_obj(bs_list, bs_obj);
b338082b 1931 }
d15e5465
LC		 1932
1933 *ret_data = QOBJECT(bs_list);
b338082b 1934}
a36e69dd 1935
218a536a 1936static void bdrv_stats_iter(QObject *data, void *opaque)
a36e69dd 1937{
218a536a
LC		 1938 QDict *qdict;
1939 Monitor *mon = opaque;
1940
1941 qdict = qobject_to_qdict(data);
1942 monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1943
1944 qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1945 monitor_printf(mon, " rd_bytes=%" PRId64
1946 " wr_bytes=%" PRId64
1947 " rd_operations=%" PRId64
1948 " wr_operations=%" PRId64
e8045d67 1949 " flush_operations=%" PRId64
c488c7f6
CH		 1950 " wr_total_time_ns=%" PRId64
1951 " rd_total_time_ns=%" PRId64
1952 " flush_total_time_ns=%" PRId64
218a536a
LC		 1953 "\n",
1954 qdict_get_int(qdict, "rd_bytes"),
1955 qdict_get_int(qdict, "wr_bytes"),
1956 qdict_get_int(qdict, "rd_operations"),
e8045d67 1957 qdict_get_int(qdict, "wr_operations"),
c488c7f6
CH		 1958 qdict_get_int(qdict, "flush_operations"),
1959 qdict_get_int(qdict, "wr_total_time_ns"),
1960 qdict_get_int(qdict, "rd_total_time_ns"),
1961 qdict_get_int(qdict, "flush_total_time_ns"));
218a536a
LC		 1962}
1963
1964void bdrv_stats_print(Monitor *mon, const QObject *data)
1965{
1966 qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1967}
1968
294cc35f
KW		 1969static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1970{
1971 QObject *res;
1972 QDict *dict;
1973
1974 res = qobject_from_jsonf("{ 'stats': {"
1975 "'rd_bytes': %" PRId64 ","
1976 "'wr_bytes': %" PRId64 ","
1977 "'rd_operations': %" PRId64 ","
1978 "'wr_operations': %" PRId64 ","
e8045d67 1979 "'wr_highest_offset': %" PRId64 ","
c488c7f6
CH		 1980 "'flush_operations': %" PRId64 ","
1981 "'wr_total_time_ns': %" PRId64 ","
1982 "'rd_total_time_ns': %" PRId64 ","
1983 "'flush_total_time_ns': %" PRId64
294cc35f 1984 "} }",
a597e79c
CH		 1985 bs->nr_bytes[BDRV_ACCT_READ],
1986 bs->nr_bytes[BDRV_ACCT_WRITE],
1987 bs->nr_ops[BDRV_ACCT_READ],
1988 bs->nr_ops[BDRV_ACCT_WRITE],
5ffbbc67 1989 bs->wr_highest_sector *
e8045d67 1990 (uint64_t)BDRV_SECTOR_SIZE,
c488c7f6
CH		 1991 bs->nr_ops[BDRV_ACCT_FLUSH],
1992 bs->total_time_ns[BDRV_ACCT_WRITE],
1993 bs->total_time_ns[BDRV_ACCT_READ],
1994 bs->total_time_ns[BDRV_ACCT_FLUSH]);
294cc35f
KW		 1995 dict = qobject_to_qdict(res);
1996
1997 if (*bs->device_name) {
1998 qdict_put(dict, "device", qstring_from_str(bs->device_name));
1999 }
2000
2001 if (bs->file) {
2002 QObject *parent = bdrv_info_stats_bs(bs->file);
2003 qdict_put_obj(dict, "parent", parent);
2004 }
2005
2006 return res;
2007}
2008
218a536a
LC		 2009void bdrv_info_stats(Monitor *mon, QObject **ret_data)
2010{
2011 QObject *obj;
2012 QList *devices;
a36e69dd
TS		 2013 BlockDriverState *bs;
2014
218a536a
LC		 2015 devices = qlist_new();
2016
1b7bdbc1 2017 QTAILQ_FOREACH(bs, &bdrv_states, list) {
294cc35f 2018 obj = bdrv_info_stats_bs(bs);
218a536a 2019 qlist_append_obj(devices, obj);
a36e69dd 2020 }
218a536a
LC		 2021
2022 *ret_data = QOBJECT(devices);
a36e69dd 2023}
ea2384d3 2024
045df330
AL		 2025const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2026{
2027 if (bs->backing_hd && bs->backing_hd->encrypted)
2028 return bs->backing_file;
2029 else if (bs->encrypted)
2030 return bs->filename;
2031 else
2032 return NULL;
2033}
2034
5fafdf24 2035void bdrv_get_backing_filename(BlockDriverState *bs,
83f64091
FB		 2036 char *filename, int filename_size)
2037{
b783e409 2038 if (!bs->backing_file) {
83f64091
FB		 2039 pstrcpy(filename, filename_size, "");
2040 } else {
2041 pstrcpy(filename, filename_size, bs->backing_file);
2042 }
2043}
2044
5fafdf24 2045int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
faea38e7
FB		 2046 const uint8_t *buf, int nb_sectors)
2047{
2048 BlockDriver *drv = bs->drv;
2049 if (!drv)
19cb3738 2050 return -ENOMEDIUM;
faea38e7
FB		 2051 if (!drv->bdrv_write_compressed)
2052 return -ENOTSUP;
fbb7b4e0
KW		 2053 if (bdrv_check_request(bs, sector_num, nb_sectors))
2054 return -EIO;
a55eb92c 2055
c6d22830 2056 if (bs->dirty_bitmap) {
7cd1e32a
LS		 2057 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2058 }
a55eb92c 2059
faea38e7
FB		 2060 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2061}
3b46e624 2062
faea38e7
FB		 2063int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2064{
2065 BlockDriver *drv = bs->drv;
2066 if (!drv)
19cb3738 2067 return -ENOMEDIUM;
faea38e7
FB		 2068 if (!drv->bdrv_get_info)
2069 return -ENOTSUP;
2070 memset(bdi, 0, sizeof(*bdi));
2071 return drv->bdrv_get_info(bs, bdi);
2072}
2073
45566e9c
CH		 2074int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2075 int64_t pos, int size)
178e08a5
AL		 2076{
2077 BlockDriver *drv = bs->drv;
2078 if (!drv)
2079 return -ENOMEDIUM;
7cdb1f6d
MK		 2080 if (drv->bdrv_save_vmstate)
2081 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2082 if (bs->file)
2083 return bdrv_save_vmstate(bs->file, buf, pos, size);
2084 return -ENOTSUP;
178e08a5
AL		 2085}
2086
45566e9c
CH		 2087int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2088 int64_t pos, int size)
178e08a5
AL		 2089{
2090 BlockDriver *drv = bs->drv;
2091 if (!drv)
2092 return -ENOMEDIUM;
7cdb1f6d
MK		 2093 if (drv->bdrv_load_vmstate)
2094 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2095 if (bs->file)
2096 return bdrv_load_vmstate(bs->file, buf, pos, size);
2097 return -ENOTSUP;
178e08a5
AL		 2098}
2099
8b9b0cc2
KW		 2100void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2101{
2102 BlockDriver *drv = bs->drv;
2103
2104 if (!drv || !drv->bdrv_debug_event) {
2105 return;
2106 }
2107
2108 return drv->bdrv_debug_event(bs, event);
2109
2110}
2111
faea38e7
FB		 2112/**************************************************************/
2113/* handling of snapshots */
2114
feeee5ac
MDCF		 2115int bdrv_can_snapshot(BlockDriverState *bs)
2116{
2117 BlockDriver *drv = bs->drv;
07b70bfb 2118 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
feeee5ac
MDCF		 2119 return 0;
2120 }
2121
2122 if (!drv->bdrv_snapshot_create) {
2123 if (bs->file != NULL) {
2124 return bdrv_can_snapshot(bs->file);
2125 }
2126 return 0;
2127 }
2128
2129 return 1;
2130}
2131
199630b6
BS		 2132int bdrv_is_snapshot(BlockDriverState *bs)
2133{
2134 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2135}
2136
f9092b10
MA		 2137BlockDriverState *bdrv_snapshots(void)
2138{
2139 BlockDriverState *bs;
2140
3ac906f7 2141 if (bs_snapshots) {
f9092b10 2142 return bs_snapshots;
3ac906f7 2143 }
f9092b10
MA		 2144
2145 bs = NULL;
2146 while ((bs = bdrv_next(bs))) {
2147 if (bdrv_can_snapshot(bs)) {
3ac906f7
MA		 2148 bs_snapshots = bs;
2149 return bs;
f9092b10
MA		 2150 }
2151 }
2152 return NULL;
f9092b10
MA		 2153}
2154
5fafdf24 2155int bdrv_snapshot_create(BlockDriverState *bs,
faea38e7
FB		 2156 QEMUSnapshotInfo *sn_info)
2157{
2158 BlockDriver *drv = bs->drv;
2159 if (!drv)
19cb3738 2160 return -ENOMEDIUM;
7cdb1f6d
MK		 2161 if (drv->bdrv_snapshot_create)
2162 return drv->bdrv_snapshot_create(bs, sn_info);
2163 if (bs->file)
2164 return bdrv_snapshot_create(bs->file, sn_info);
2165 return -ENOTSUP;
faea38e7
FB		 2166}
2167
5fafdf24 2168int bdrv_snapshot_goto(BlockDriverState *bs,
faea38e7
FB		 2169 const char *snapshot_id)
2170{
2171 BlockDriver *drv = bs->drv;
7cdb1f6d
MK		 2172 int ret, open_ret;
2173
faea38e7 2174 if (!drv)
19cb3738 2175 return -ENOMEDIUM;
7cdb1f6d
MK		 2176 if (drv->bdrv_snapshot_goto)
2177 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2178
2179 if (bs->file) {
2180 drv->bdrv_close(bs);
2181 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2182 open_ret = drv->bdrv_open(bs, bs->open_flags);
2183 if (open_ret < 0) {
2184 bdrv_delete(bs->file);
2185 bs->drv = NULL;
2186 return open_ret;
2187 }
2188 return ret;
2189 }
2190
2191 return -ENOTSUP;
faea38e7
FB		 2192}
2193
2194int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2195{
2196 BlockDriver *drv = bs->drv;
2197 if (!drv)
19cb3738 2198 return -ENOMEDIUM;
7cdb1f6d
MK		 2199 if (drv->bdrv_snapshot_delete)
2200 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2201 if (bs->file)
2202 return bdrv_snapshot_delete(bs->file, snapshot_id);
2203 return -ENOTSUP;
faea38e7
FB		 2204}
2205
5fafdf24 2206int bdrv_snapshot_list(BlockDriverState *bs,
faea38e7
FB		 2207 QEMUSnapshotInfo **psn_info)
2208{
2209 BlockDriver *drv = bs->drv;
2210 if (!drv)
19cb3738 2211 return -ENOMEDIUM;
7cdb1f6d
MK		 2212 if (drv->bdrv_snapshot_list)
2213 return drv->bdrv_snapshot_list(bs, psn_info);
2214 if (bs->file)
2215 return bdrv_snapshot_list(bs->file, psn_info);
2216 return -ENOTSUP;
faea38e7
FB		 2217}
2218
51ef6727 2219int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2220 const char *snapshot_name)
2221{
2222 BlockDriver *drv = bs->drv;
2223 if (!drv) {
2224 return -ENOMEDIUM;
2225 }
2226 if (!bs->read_only) {
2227 return -EINVAL;
2228 }
2229 if (drv->bdrv_snapshot_load_tmp) {
2230 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2231 }
2232 return -ENOTSUP;
2233}
2234
faea38e7
FB		 2235#define NB_SUFFIXES 4
2236
2237char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2238{
2239 static const char suffixes[NB_SUFFIXES] = "KMGT";
2240 int64_t base;
2241 int i;
2242
2243 if (size <= 999) {
2244 snprintf(buf, buf_size, "%" PRId64, size);
2245 } else {
2246 base = 1024;
2247 for(i = 0; i < NB_SUFFIXES; i++) {
2248 if (size < (10 * base)) {
5fafdf24 2249 snprintf(buf, buf_size, "%0.1f%c",
faea38e7
FB		 2250 (double)size / base,
2251 suffixes[i]);
2252 break;
2253 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
5fafdf24 2254 snprintf(buf, buf_size, "%" PRId64 "%c",
faea38e7
FB		 2255 ((size + (base >> 1)) / base),
2256 suffixes[i]);
2257 break;
2258 }
2259 base = base * 1024;
2260 }
2261 }
2262 return buf;
2263}
2264
2265char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2266{
2267 char buf1[128], date_buf[128], clock_buf[128];
3b9f94e1
FB		 2268#ifdef _WIN32
2269 struct tm *ptm;
2270#else
faea38e7 2271 struct tm tm;
3b9f94e1 2272#endif
faea38e7
FB		 2273 time_t ti;
2274 int64_t secs;
2275
2276 if (!sn) {
5fafdf24
TS		 2277 snprintf(buf, buf_size,
2278 "%-10s%-20s%7s%20s%15s",
faea38e7
FB		 2279 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2280 } else {
2281 ti = sn->date_sec;
3b9f94e1
FB		 2282#ifdef _WIN32
2283 ptm = localtime(&ti);
2284 strftime(date_buf, sizeof(date_buf),
2285 "%Y-%m-%d %H:%M:%S", ptm);
2286#else
faea38e7
FB		 2287 localtime_r(&ti, &tm);
2288 strftime(date_buf, sizeof(date_buf),
2289 "%Y-%m-%d %H:%M:%S", &tm);
3b9f94e1 2290#endif
faea38e7
FB		 2291 secs = sn->vm_clock_nsec / 1000000000;
2292 snprintf(clock_buf, sizeof(clock_buf),
2293 "%02d:%02d:%02d.%03d",
2294 (int)(secs / 3600),
2295 (int)((secs / 60) % 60),
5fafdf24 2296 (int)(secs % 60),
faea38e7
FB		 2297 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2298 snprintf(buf, buf_size,
5fafdf24 2299 "%-10s%-20s%7s%20s%15s",
faea38e7
FB		 2300 sn->id_str, sn->name,
2301 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2302 date_buf,
2303 clock_buf);
2304 }
2305 return buf;
2306}
2307
ea2384d3 2308/**************************************************************/
83f64091 2309/* async I/Os */
ea2384d3 2310
3b69e4b9 2311BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
f141eafe 2312 QEMUIOVector *qiov, int nb_sectors,
3b69e4b9 2313 BlockDriverCompletionFunc *cb, void *opaque)
83f64091
FB		 2314{
2315 BlockDriver *drv = bs->drv;
83f64091 2316
bbf0a440
SH		 2317 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2318
19cb3738 2319 if (!drv)
ce1a14dc 2320 return NULL;
71d0770c
AL		 2321 if (bdrv_check_request(bs, sector_num, nb_sectors))
2322 return NULL;
3b46e624 2323
a597e79c
CH		 2324 return drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
2325 cb, opaque);
ea2384d3
FB		 2326}
2327
4dcafbb1
MT		 2328typedef struct BlockCompleteData {
2329 BlockDriverCompletionFunc *cb;
2330 void *opaque;
2331 BlockDriverState *bs;
2332 int64_t sector_num;
2333 int nb_sectors;
2334} BlockCompleteData;
2335
2336static void block_complete_cb(void *opaque, int ret)
2337{
2338 BlockCompleteData *b = opaque;
2339
2340 if (b->bs->dirty_bitmap) {
2341 set_dirty_bitmap(b->bs, b->sector_num, b->nb_sectors, 1);
2342 }
2343 b->cb(b->opaque, ret);
7267c094 2344 g_free(b);
4dcafbb1
MT		 2345}
2346
2347static BlockCompleteData *blk_dirty_cb_alloc(BlockDriverState *bs,
2348 int64_t sector_num,
2349 int nb_sectors,
2350 BlockDriverCompletionFunc *cb,
2351 void *opaque)
2352{
7267c094 2353 BlockCompleteData *blkdata = g_malloc0(sizeof(BlockCompleteData));
4dcafbb1
MT		 2354
2355 blkdata->bs = bs;
2356 blkdata->cb = cb;
2357 blkdata->opaque = opaque;
2358 blkdata->sector_num = sector_num;
2359 blkdata->nb_sectors = nb_sectors;
2360
2361 return blkdata;
2362}
2363
f141eafe
AL		 2364BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2365 QEMUIOVector *qiov, int nb_sectors,
2366 BlockDriverCompletionFunc *cb, void *opaque)
ea2384d3 2367{
83f64091 2368 BlockDriver *drv = bs->drv;
a36e69dd 2369 BlockDriverAIOCB *ret;
4dcafbb1 2370 BlockCompleteData *blk_cb_data;
ea2384d3 2371
bbf0a440
SH		 2372 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2373
19cb3738 2374 if (!drv)
ce1a14dc 2375 return NULL;
83f64091 2376 if (bs->read_only)
ce1a14dc 2377 return NULL;
71d0770c
AL		 2378 if (bdrv_check_request(bs, sector_num, nb_sectors))
2379 return NULL;
83f64091 2380
c6d22830 2381 if (bs->dirty_bitmap) {
4dcafbb1
MT		 2382 blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
2383 opaque);
2384 cb = &block_complete_cb;
2385 opaque = blk_cb_data;
7cd1e32a 2386 }
a55eb92c 2387
f141eafe
AL		 2388 ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2389 cb, opaque);
a36e69dd
TS		 2390
2391 if (ret) {
294cc35f
KW		 2392 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2393 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2394 }
a36e69dd
TS		 2395 }
2396
2397 return ret;
83f64091
FB		 2398}
2399
40b4f539
KW		 2400
2401typedef struct MultiwriteCB {
2402 int error;
2403 int num_requests;
2404 int num_callbacks;
2405 struct {
2406 BlockDriverCompletionFunc *cb;
2407 void *opaque;
2408 QEMUIOVector *free_qiov;
2409 void *free_buf;
2410 } callbacks[];
2411} MultiwriteCB;
2412
2413static void multiwrite_user_cb(MultiwriteCB *mcb)
2414{
2415 int i;
2416
2417 for (i = 0; i < mcb->num_callbacks; i++) {
2418 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1e1ea48d
SH		 2419 if (mcb->callbacks[i].free_qiov) {
2420 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2421 }
7267c094 2422 g_free(mcb->callbacks[i].free_qiov);
f8a83245 2423 qemu_vfree(mcb->callbacks[i].free_buf);
40b4f539
KW		 2424 }
2425}
2426
2427static void multiwrite_cb(void *opaque, int ret)
2428{
2429 MultiwriteCB *mcb = opaque;
2430
6d519a5f
SH		 2431 trace_multiwrite_cb(mcb, ret);
2432
cb6d3ca0 2433 if (ret < 0 && !mcb->error) {
40b4f539 2434 mcb->error = ret;
40b4f539
KW		 2435 }
2436
2437 mcb->num_requests--;
2438 if (mcb->num_requests == 0) {
de189a1b 2439 multiwrite_user_cb(mcb);
7267c094 2440 g_free(mcb);
40b4f539
KW		 2441 }
2442}
2443
2444static int multiwrite_req_compare(const void *a, const void *b)
2445{
77be4366
CH		 2446 const BlockRequest *req1 = a, *req2 = b;
2447
2448 /*
2449 * Note that we can't simply subtract req2->sector from req1->sector
2450 * here as that could overflow the return value.
2451 */
2452 if (req1->sector > req2->sector) {
2453 return 1;
2454 } else if (req1->sector < req2->sector) {
2455 return -1;
2456 } else {
2457 return 0;
2458 }
40b4f539
KW		 2459}
2460
2461/*
2462 * Takes a bunch of requests and tries to merge them. Returns the number of
2463 * requests that remain after merging.
2464 */
2465static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2466 int num_reqs, MultiwriteCB *mcb)
2467{
2468 int i, outidx;
2469
2470 // Sort requests by start sector
2471 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2472
2473 // Check if adjacent requests touch the same clusters. If so, combine them,
2474 // filling up gaps with zero sectors.
2475 outidx = 0;
2476 for (i = 1; i < num_reqs; i++) {
2477 int merge = 0;
2478 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2479
2480 // This handles the cases that are valid for all block drivers, namely
2481 // exactly sequential writes and overlapping writes.
2482 if (reqs[i].sector <= oldreq_last) {
2483 merge = 1;
2484 }
2485
2486 // The block driver may decide that it makes sense to combine requests
2487 // even if there is a gap of some sectors between them. In this case,
2488 // the gap is filled with zeros (therefore only applicable for yet
2489 // unused space in format like qcow2).
2490 if (!merge && bs->drv->bdrv_merge_requests) {
2491 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2492 }
2493
e2a305fb
CH		 2494 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2495 merge = 0;
2496 }
2497
40b4f539
KW		 2498 if (merge) {
2499 size_t size;
7267c094 2500 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
40b4f539
KW		 2501 qemu_iovec_init(qiov,
2502 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2503
2504 // Add the first request to the merged one. If the requests are
2505 // overlapping, drop the last sectors of the first request.
2506 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2507 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2508
2509 // We might need to add some zeros between the two requests
2510 if (reqs[i].sector > oldreq_last) {
2511 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2512 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2513 memset(buf, 0, zero_bytes);
2514 qemu_iovec_add(qiov, buf, zero_bytes);
2515 mcb->callbacks[i].free_buf = buf;
2516 }
2517
2518 // Add the second request
2519 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2520
cbf1dff2 2521 reqs[outidx].nb_sectors = qiov->size >> 9;
40b4f539
KW		 2522 reqs[outidx].qiov = qiov;
2523
2524 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2525 } else {
2526 outidx++;
2527 reqs[outidx].sector = reqs[i].sector;
2528 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2529 reqs[outidx].qiov = reqs[i].qiov;
2530 }
2531 }
2532
2533 return outidx + 1;
2534}
2535
2536/*
2537 * Submit multiple AIO write requests at once.
2538 *
2539 * On success, the function returns 0 and all requests in the reqs array have
2540 * been submitted. In error case this function returns -1, and any of the
2541 * requests may or may not be submitted yet. In particular, this means that the
2542 * callback will be called for some of the requests, for others it won't. The
2543 * caller must check the error field of the BlockRequest to wait for the right
2544 * callbacks (if error != 0, no callback will be called).
2545 *
2546 * The implementation may modify the contents of the reqs array, e.g. to merge
2547 * requests. However, the fields opaque and error are left unmodified as they
2548 * are used to signal failure for a single request to the caller.
2549 */
2550int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2551{
2552 BlockDriverAIOCB *acb;
2553 MultiwriteCB *mcb;
2554 int i;
2555
301db7c2
RH		 2556 /* don't submit writes if we don't have a medium */
2557 if (bs->drv == NULL) {
2558 for (i = 0; i < num_reqs; i++) {
2559 reqs[i].error = -ENOMEDIUM;
2560 }
2561 return -1;
2562 }
2563
40b4f539
KW		 2564 if (num_reqs == 0) {
2565 return 0;
2566 }
2567
2568 // Create MultiwriteCB structure
7267c094 2569 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
40b4f539
KW		 2570 mcb->num_requests = 0;
2571 mcb->num_callbacks = num_reqs;
2572
2573 for (i = 0; i < num_reqs; i++) {
2574 mcb->callbacks[i].cb = reqs[i].cb;
2575 mcb->callbacks[i].opaque = reqs[i].opaque;
2576 }
2577
2578 // Check for mergable requests
2579 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2580
6d519a5f
SH		 2581 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2582
453f9a16
KW		 2583 /*
2584 * Run the aio requests. As soon as one request can't be submitted
2585 * successfully, fail all requests that are not yet submitted (we must
2586 * return failure for all requests anyway)
2587 *
2588 * num_requests cannot be set to the right value immediately: If
2589 * bdrv_aio_writev fails for some request, num_requests would be too high
2590 * and therefore multiwrite_cb() would never recognize the multiwrite
2591 * request as completed. We also cannot use the loop variable i to set it
2592 * when the first request fails because the callback may already have been
2593 * called for previously submitted requests. Thus, num_requests must be
2594 * incremented for each request that is submitted.
2595 *
2596 * The problem that callbacks may be called early also means that we need
2597 * to take care that num_requests doesn't become 0 before all requests are
2598 * submitted - multiwrite_cb() would consider the multiwrite request
2599 * completed. A dummy request that is "completed" by a manual call to
2600 * multiwrite_cb() takes care of this.
2601 */
2602 mcb->num_requests = 1;
2603
6d519a5f 2604 // Run the aio requests
40b4f539 2605 for (i = 0; i < num_reqs; i++) {
453f9a16 2606 mcb->num_requests++;
40b4f539
KW		 2607 acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2608 reqs[i].nb_sectors, multiwrite_cb, mcb);
2609
2610 if (acb == NULL) {
2611 // We can only fail the whole thing if no request has been
2612 // submitted yet. Otherwise we'll wait for the submitted AIOs to
2613 // complete and report the error in the callback.
453f9a16 2614 if (i == 0) {
6d519a5f 2615 trace_bdrv_aio_multiwrite_earlyfail(mcb);
40b4f539
KW		 2616 goto fail;
2617 } else {
6d519a5f 2618 trace_bdrv_aio_multiwrite_latefail(mcb, i);
7eb58a6c 2619 multiwrite_cb(mcb, -EIO);
40b4f539
KW		 2620 break;
2621 }
40b4f539
KW		 2622 }
2623 }
2624
453f9a16
KW		 2625 /* Complete the dummy request */
2626 multiwrite_cb(mcb, 0);
2627
40b4f539
KW		 2628 return 0;
2629
2630fail:
453f9a16
KW		 2631 for (i = 0; i < mcb->num_callbacks; i++) {
2632 reqs[i].error = -EIO;
2633 }
7267c094 2634 g_free(mcb);
40b4f539
KW		 2635 return -1;
2636}
2637
b2e12bc6
CH		 2638BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2639 BlockDriverCompletionFunc *cb, void *opaque)
2640{
2641 BlockDriver *drv = bs->drv;
2642
a13aac04
SH		 2643 trace_bdrv_aio_flush(bs, opaque);
2644
016f5cf6
AG		 2645 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2646 return bdrv_aio_noop_em(bs, cb, opaque);
2647 }
2648
b2e12bc6
CH		 2649 if (!drv)
2650 return NULL;
b2e12bc6
CH		 2651 return drv->bdrv_aio_flush(bs, cb, opaque);
2652}
2653
83f64091 2654void bdrv_aio_cancel(BlockDriverAIOCB *acb)
83f64091 2655{
6bbff9a0 2656 acb->pool->cancel(acb);
83f64091
FB		 2657}
2658
ce1a14dc 2659
83f64091
FB		 2660/**************************************************************/
2661/* async block device emulation */
2662
c16b5a2c
CH		 2663typedef struct BlockDriverAIOCBSync {
2664 BlockDriverAIOCB common;
2665 QEMUBH *bh;
2666 int ret;
2667 /* vector translation state */
2668 QEMUIOVector *qiov;
2669 uint8_t *bounce;
2670 int is_write;
2671} BlockDriverAIOCBSync;
2672
2673static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2674{
b666d239
KW		 2675 BlockDriverAIOCBSync *acb =
2676 container_of(blockacb, BlockDriverAIOCBSync, common);
6a7ad299 2677 qemu_bh_delete(acb->bh);
36afc451 2678 acb->bh = NULL;
c16b5a2c
CH		 2679 qemu_aio_release(acb);
2680}
2681
2682static AIOPool bdrv_em_aio_pool = {
2683 .aiocb_size = sizeof(BlockDriverAIOCBSync),
2684 .cancel = bdrv_aio_cancel_em,
2685};
2686
ce1a14dc 2687static void bdrv_aio_bh_cb(void *opaque)
83f64091 2688{
ce1a14dc 2689 BlockDriverAIOCBSync *acb = opaque;
f141eafe 2690
f141eafe
AL		 2691 if (!acb->is_write)
2692 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
ceb42de8 2693 qemu_vfree(acb->bounce);
ce1a14dc 2694 acb->common.cb(acb->common.opaque, acb->ret);
6a7ad299 2695 qemu_bh_delete(acb->bh);
36afc451 2696 acb->bh = NULL;
ce1a14dc 2697 qemu_aio_release(acb);
83f64091 2698}
beac80cd 2699
f141eafe
AL		 2700static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2701 int64_t sector_num,
2702 QEMUIOVector *qiov,
2703 int nb_sectors,
2704 BlockDriverCompletionFunc *cb,
2705 void *opaque,
2706 int is_write)
2707
83f64091 2708{
ce1a14dc 2709 BlockDriverAIOCBSync *acb;
ce1a14dc 2710
c16b5a2c 2711 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
f141eafe
AL		 2712 acb->is_write = is_write;
2713 acb->qiov = qiov;
e268ca52 2714 acb->bounce = qemu_blockalign(bs, qiov->size);
f141eafe 2715
ce1a14dc
PB		 2716 if (!acb->bh)
2717 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
f141eafe
AL		 2718
2719 if (is_write) {
2720 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2721 acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2722 } else {
2723 acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2724 }
2725
ce1a14dc 2726 qemu_bh_schedule(acb->bh);
f141eafe 2727
ce1a14dc 2728 return &acb->common;
beac80cd
FB		 2729}
2730
f141eafe
AL		 2731static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2732 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc 2733 BlockDriverCompletionFunc *cb, void *opaque)
beac80cd 2734{
f141eafe
AL		 2735 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2736}
83f64091 2737
f141eafe
AL		 2738static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2739 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2740 BlockDriverCompletionFunc *cb, void *opaque)
2741{
2742 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
beac80cd 2743}
beac80cd 2744
68485420
KW		 2745
2746typedef struct BlockDriverAIOCBCoroutine {
2747 BlockDriverAIOCB common;
2748 BlockRequest req;
2749 bool is_write;
2750 QEMUBH* bh;
2751} BlockDriverAIOCBCoroutine;
2752
2753static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
2754{
2755 qemu_aio_flush();
2756}
2757
2758static AIOPool bdrv_em_co_aio_pool = {
2759 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
2760 .cancel = bdrv_aio_co_cancel_em,
2761};
2762
2763static void bdrv_co_rw_bh(void *opaque)
2764{
2765 BlockDriverAIOCBCoroutine *acb = opaque;
2766
2767 acb->common.cb(acb->common.opaque, acb->req.error);
2768 qemu_bh_delete(acb->bh);
2769 qemu_aio_release(acb);
2770}
2771
2772static void coroutine_fn bdrv_co_rw(void *opaque)
2773{
2774 BlockDriverAIOCBCoroutine *acb = opaque;
2775 BlockDriverState *bs = acb->common.bs;
2776
2777 if (!acb->is_write) {
2778 acb->req.error = bs->drv->bdrv_co_readv(bs, acb->req.sector,
2779 acb->req.nb_sectors, acb->req.qiov);
2780 } else {
2781 acb->req.error = bs->drv->bdrv_co_writev(bs, acb->req.sector,
2782 acb->req.nb_sectors, acb->req.qiov);
2783 }
2784
2785 acb->bh = qemu_bh_new(bdrv_co_rw_bh, acb);
2786 qemu_bh_schedule(acb->bh);
2787}
2788
2789static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
2790 int64_t sector_num,
2791 QEMUIOVector *qiov,
2792 int nb_sectors,
2793 BlockDriverCompletionFunc *cb,
2794 void *opaque,
2795 bool is_write)
2796{
2797 Coroutine *co;
2798 BlockDriverAIOCBCoroutine *acb;
2799
2800 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
2801 acb->req.sector = sector_num;
2802 acb->req.nb_sectors = nb_sectors;
2803 acb->req.qiov = qiov;
2804 acb->is_write = is_write;
2805
2806 co = qemu_coroutine_create(bdrv_co_rw);
2807 qemu_coroutine_enter(co, acb);
2808
2809 return &acb->common;
2810}
2811
2812static BlockDriverAIOCB *bdrv_co_aio_readv_em(BlockDriverState *bs,
2813 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2814 BlockDriverCompletionFunc *cb, void *opaque)
2815{
2816 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque,
2817 false);
2818}
2819
2820static BlockDriverAIOCB *bdrv_co_aio_writev_em(BlockDriverState *bs,
2821 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2822 BlockDriverCompletionFunc *cb, void *opaque)
2823{
2824 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque,
2825 true);
2826}
2827
b2e12bc6
CH		 2828static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2829 BlockDriverCompletionFunc *cb, void *opaque)
2830{
2831 BlockDriverAIOCBSync *acb;
2832
2833 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2834 acb->is_write = 1; /* don't bounce in the completion hadler */
2835 acb->qiov = NULL;
2836 acb->bounce = NULL;
2837 acb->ret = 0;
2838
2839 if (!acb->bh)
2840 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2841
2842 bdrv_flush(bs);
2843 qemu_bh_schedule(acb->bh);
2844 return &acb->common;
2845}
2846
016f5cf6
AG		 2847static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2848 BlockDriverCompletionFunc *cb, void *opaque)
2849{
2850 BlockDriverAIOCBSync *acb;
2851
2852 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2853 acb->is_write = 1; /* don't bounce in the completion handler */
2854 acb->qiov = NULL;
2855 acb->bounce = NULL;
2856 acb->ret = 0;
2857
2858 if (!acb->bh) {
2859 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2860 }
2861
2862 qemu_bh_schedule(acb->bh);
2863 return &acb->common;
2864}
2865
83f64091
FB		 2866/**************************************************************/
2867/* sync block device emulation */
ea2384d3 2868
83f64091
FB		 2869static void bdrv_rw_em_cb(void *opaque, int ret)
2870{
2871 *(int *)opaque = ret;
ea2384d3
FB		 2872}
2873
83f64091
FB		 2874#define NOT_DONE 0x7fffffff
2875
5fafdf24 2876static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
83f64091 2877 uint8_t *buf, int nb_sectors)
7a6cba61 2878{
ce1a14dc
PB		 2879 int async_ret;
2880 BlockDriverAIOCB *acb;
f141eafe
AL		 2881 struct iovec iov;
2882 QEMUIOVector qiov;
83f64091 2883
83f64091 2884 async_ret = NOT_DONE;
3f4cb3d3 2885 iov.iov_base = (void *)buf;
eb5a3165 2886 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
f141eafe
AL		 2887 qemu_iovec_init_external(&qiov, &iov, 1);
2888 acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2889 bdrv_rw_em_cb, &async_ret);
65d6b3d8
KW		 2890 if (acb == NULL) {
2891 async_ret = -1;
2892 goto fail;
2893 }
baf35cb9 2894
83f64091
FB		 2895 while (async_ret == NOT_DONE) {
2896 qemu_aio_wait();
2897 }
baf35cb9 2898
65d6b3d8
KW		 2899
2900fail:
83f64091 2901 return async_ret;
7a6cba61
PB		 2902}
2903
83f64091
FB		 2904static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2905 const uint8_t *buf, int nb_sectors)
2906{
ce1a14dc
PB		 2907 int async_ret;
2908 BlockDriverAIOCB *acb;
f141eafe
AL		 2909 struct iovec iov;
2910 QEMUIOVector qiov;
83f64091 2911
83f64091 2912 async_ret = NOT_DONE;
f141eafe 2913 iov.iov_base = (void *)buf;
eb5a3165 2914 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
f141eafe
AL		 2915 qemu_iovec_init_external(&qiov, &iov, 1);
2916 acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2917 bdrv_rw_em_cb, &async_ret);
65d6b3d8
KW		 2918 if (acb == NULL) {
2919 async_ret = -1;
2920 goto fail;
2921 }
83f64091
FB		 2922 while (async_ret == NOT_DONE) {
2923 qemu_aio_wait();
2924 }
65d6b3d8
KW		 2925
2926fail:
83f64091
FB		 2927 return async_ret;
2928}
ea2384d3
FB		 2929
2930void bdrv_init(void)
2931{
5efa9d5a 2932 module_call_init(MODULE_INIT_BLOCK);
ea2384d3 2933}
ce1a14dc 2934
eb852011
MA		 2935void bdrv_init_with_whitelist(void)
2936{
2937 use_bdrv_whitelist = 1;
2938 bdrv_init();
2939}
2940
c16b5a2c
CH		 2941void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2942 BlockDriverCompletionFunc *cb, void *opaque)
ce1a14dc 2943{
ce1a14dc
PB		 2944 BlockDriverAIOCB *acb;
2945
6bbff9a0
AL		 2946 if (pool->free_aiocb) {
2947 acb = pool->free_aiocb;
2948 pool->free_aiocb = acb->next;
ce1a14dc 2949 } else {
7267c094 2950 acb = g_malloc0(pool->aiocb_size);
6bbff9a0 2951 acb->pool = pool;
ce1a14dc
PB		 2952 }
2953 acb->bs = bs;
2954 acb->cb = cb;
2955 acb->opaque = opaque;
2956 return acb;
2957}
2958
2959void qemu_aio_release(void *p)
2960{
6bbff9a0
AL		 2961 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2962 AIOPool *pool = acb->pool;
2963 acb->next = pool->free_aiocb;
2964 pool->free_aiocb = acb;
ce1a14dc 2965}
19cb3738 2966
f9f05dc5
KW		 2967/**************************************************************/
2968/* Coroutine block device emulation */
2969
2970typedef struct CoroutineIOCompletion {
2971 Coroutine *coroutine;
2972 int ret;
2973} CoroutineIOCompletion;
2974
2975static void bdrv_co_io_em_complete(void *opaque, int ret)
2976{
2977 CoroutineIOCompletion *co = opaque;
2978
2979 co->ret = ret;
2980 qemu_coroutine_enter(co->coroutine, NULL);
2981}
2982
2983static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
2984 int nb_sectors, QEMUIOVector *iov,
2985 bool is_write)
2986{
2987 CoroutineIOCompletion co = {
2988 .coroutine = qemu_coroutine_self(),
2989 };
2990 BlockDriverAIOCB *acb;
2991
2992 if (is_write) {
2993 acb = bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
2994 bdrv_co_io_em_complete, &co);
2995 } else {
2996 acb = bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
2997 bdrv_co_io_em_complete, &co);
2998 }
2999
3000 trace_bdrv_co_io(is_write, acb);
3001 if (!acb) {
3002 return -EIO;
3003 }
3004 qemu_coroutine_yield();
3005
3006 return co.ret;
3007}
3008
3009static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3010 int64_t sector_num, int nb_sectors,
3011 QEMUIOVector *iov)
3012{
3013 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3014}
3015
3016static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3017 int64_t sector_num, int nb_sectors,
3018 QEMUIOVector *iov)
3019{
3020 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3021}
3022
e7a8a783
KW		 3023static int coroutine_fn bdrv_co_flush_em(BlockDriverState *bs)
3024{
3025 CoroutineIOCompletion co = {
3026 .coroutine = qemu_coroutine_self(),
3027 };
3028 BlockDriverAIOCB *acb;
3029
3030 acb = bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3031 if (!acb) {
3032 return -EIO;
3033 }
3034 qemu_coroutine_yield();
3035 return co.ret;
3036}
3037
19cb3738
FB		 3038/**************************************************************/
3039/* removable device support */
3040
3041/**
3042 * Return TRUE if the media is present
3043 */
3044int bdrv_is_inserted(BlockDriverState *bs)
3045{
3046 BlockDriver *drv = bs->drv;
a1aff5bf 3047
19cb3738
FB		 3048 if (!drv)
3049 return 0;
3050 if (!drv->bdrv_is_inserted)
a1aff5bf
MA		 3051 return 1;
3052 return drv->bdrv_is_inserted(bs);
19cb3738
FB		 3053}
3054
3055/**
8e49ca46
MA		 3056 * Return whether the media changed since the last call to this
3057 * function, or -ENOTSUP if we don't know. Most drivers don't know.
19cb3738
FB		 3058 */
3059int bdrv_media_changed(BlockDriverState *bs)
3060{
3061 BlockDriver *drv = bs->drv;
19cb3738 3062
8e49ca46
MA		 3063 if (drv && drv->bdrv_media_changed) {
3064 return drv->bdrv_media_changed(bs);
3065 }
3066 return -ENOTSUP;
19cb3738
FB		 3067}
3068
3069/**
3070 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3071 */
fdec4404 3072void bdrv_eject(BlockDriverState *bs, int eject_flag)
19cb3738
FB		 3073{
3074 BlockDriver *drv = bs->drv;
19cb3738 3075
822e1cd1
MA		 3076 if (drv && drv->bdrv_eject) {
3077 drv->bdrv_eject(bs, eject_flag);
19cb3738
FB		 3078 }
3079}
3080
19cb3738
FB		 3081/**
3082 * Lock or unlock the media (if it is locked, the user won't be able
3083 * to eject it manually).
3084 */
025e849a 3085void bdrv_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB		 3086{
3087 BlockDriver *drv = bs->drv;
3088
025e849a 3089 trace_bdrv_lock_medium(bs, locked);
b8c6d095 3090
025e849a
MA		 3091 if (drv && drv->bdrv_lock_medium) {
3092 drv->bdrv_lock_medium(bs, locked);
19cb3738
FB		 3093 }
3094}
985a03b0
TS		 3095
3096/* needed for generic scsi interface */
3097
3098int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3099{
3100 BlockDriver *drv = bs->drv;
3101
3102 if (drv && drv->bdrv_ioctl)
3103 return drv->bdrv_ioctl(bs, req, buf);
3104 return -ENOTSUP;
3105}
7d780669 3106
221f715d
AL		 3107BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3108 unsigned long int req, void *buf,
3109 BlockDriverCompletionFunc *cb, void *opaque)
7d780669 3110{
221f715d 3111 BlockDriver *drv = bs->drv;
7d780669 3112
221f715d
AL		 3113 if (drv && drv->bdrv_aio_ioctl)
3114 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3115 return NULL;
7d780669 3116}
e268ca52 3117
7b6f9300
MA		 3118void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3119{
3120 bs->buffer_alignment = align;
3121}
7cd1e32a 3122
e268ca52
AL		 3123void *qemu_blockalign(BlockDriverState *bs, size_t size)
3124{
3125 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3126}
7cd1e32a
LS		 3127
3128void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3129{
3130 int64_t bitmap_size;
a55eb92c 3131
aaa0eb75 3132 bs->dirty_count = 0;
a55eb92c 3133 if (enable) {
c6d22830
JK		 3134 if (!bs->dirty_bitmap) {
3135 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3136 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3137 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
a55eb92c 3138
7267c094 3139 bs->dirty_bitmap = g_malloc0(bitmap_size);
a55eb92c 3140 }
7cd1e32a 3141 } else {
c6d22830 3142 if (bs->dirty_bitmap) {
7267c094 3143 g_free(bs->dirty_bitmap);
c6d22830 3144 bs->dirty_bitmap = NULL;
a55eb92c 3145 }
7cd1e32a
LS		 3146 }
3147}
3148
3149int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3150{
6ea44308 3151 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
a55eb92c 3152
c6d22830
JK		 3153 if (bs->dirty_bitmap &&
3154 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
6d59fec1
MT		 3155 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3156 (1UL << (chunk % (sizeof(unsigned long) * 8))));
7cd1e32a
LS		 3157 } else {
3158 return 0;
3159 }
3160}
3161
a55eb92c
JK		 3162void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3163 int nr_sectors)
7cd1e32a
LS		 3164{
3165 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3166}
aaa0eb75
LS		 3167
3168int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3169{
3170 return bs->dirty_count;
3171}
f88e1a42 3172
db593f25
MT		 3173void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3174{
3175 assert(bs->in_use != in_use);
3176 bs->in_use = in_use;
3177}
3178
3179int bdrv_in_use(BlockDriverState *bs)
3180{
3181 return bs->in_use;
3182}
3183
a597e79c
CH		 3184void
3185bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3186 enum BlockAcctType type)
3187{
3188 assert(type < BDRV_MAX_IOTYPE);
3189
3190 cookie->bytes = bytes;
c488c7f6 3191 cookie->start_time_ns = get_clock();
a597e79c
CH		 3192 cookie->type = type;
3193}
3194
3195void
3196bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3197{
3198 assert(cookie->type < BDRV_MAX_IOTYPE);
3199
3200 bs->nr_bytes[cookie->type] += cookie->bytes;
3201 bs->nr_ops[cookie->type]++;
c488c7f6 3202 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
a597e79c
CH		 3203}
3204
f88e1a42
JS		 3205int bdrv_img_create(const char *filename, const char *fmt,
3206 const char *base_filename, const char *base_fmt,
3207 char *options, uint64_t img_size, int flags)
3208{
3209 QEMUOptionParameter *param = NULL, *create_options = NULL;
d220894e 3210 QEMUOptionParameter *backing_fmt, *backing_file, *size;
f88e1a42
JS		 3211 BlockDriverState *bs = NULL;
3212 BlockDriver *drv, *proto_drv;
96df67d1 3213 BlockDriver *backing_drv = NULL;
f88e1a42
JS		 3214 int ret = 0;
3215
3216 /* Find driver and parse its options */
3217 drv = bdrv_find_format(fmt);
3218 if (!drv) {
3219 error_report("Unknown file format '%s'", fmt);
4f70f249 3220 ret = -EINVAL;
f88e1a42
JS		 3221 goto out;
3222 }
3223
3224 proto_drv = bdrv_find_protocol(filename);
3225 if (!proto_drv) {
3226 error_report("Unknown protocol '%s'", filename);
4f70f249 3227 ret = -EINVAL;
f88e1a42
JS		 3228 goto out;
3229 }
3230
3231 create_options = append_option_parameters(create_options,
3232 drv->create_options);
3233 create_options = append_option_parameters(create_options,
3234 proto_drv->create_options);
3235
3236 /* Create parameter list with default values */
3237 param = parse_option_parameters("", create_options, param);
3238
3239 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3240
3241 /* Parse -o options */
3242 if (options) {
3243 param = parse_option_parameters(options, create_options, param);
3244 if (param == NULL) {
3245 error_report("Invalid options for file format '%s'.", fmt);
4f70f249 3246 ret = -EINVAL;
f88e1a42
JS		 3247 goto out;
3248 }
3249 }
3250
3251 if (base_filename) {
3252 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3253 base_filename)) {
3254 error_report("Backing file not supported for file format '%s'",
3255 fmt);
4f70f249 3256 ret = -EINVAL;
f88e1a42
JS		 3257 goto out;
3258 }
3259 }
3260
3261 if (base_fmt) {
3262 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
3263 error_report("Backing file format not supported for file "
3264 "format '%s'", fmt);
4f70f249 3265 ret = -EINVAL;
f88e1a42
JS		 3266 goto out;
3267 }
3268 }
3269
792da93a
JS		 3270 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
3271 if (backing_file && backing_file->value.s) {
3272 if (!strcmp(filename, backing_file->value.s)) {
3273 error_report("Error: Trying to create an image with the "
3274 "same filename as the backing file");
4f70f249 3275 ret = -EINVAL;
792da93a
JS		 3276 goto out;
3277 }
3278 }
3279
f88e1a42
JS		 3280 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
3281 if (backing_fmt && backing_fmt->value.s) {
96df67d1
SH		 3282 backing_drv = bdrv_find_format(backing_fmt->value.s);
3283 if (!backing_drv) {
f88e1a42
JS		 3284 error_report("Unknown backing file format '%s'",
3285 backing_fmt->value.s);
4f70f249 3286 ret = -EINVAL;
f88e1a42
JS		 3287 goto out;
3288 }
3289 }
3290
3291 // The size for the image must always be specified, with one exception:
3292 // If we are using a backing file, we can obtain the size from there
d220894e
KW		 3293 size = get_option_parameter(param, BLOCK_OPT_SIZE);
3294 if (size && size->value.n == -1) {
f88e1a42
JS		 3295 if (backing_file && backing_file->value.s) {
3296 uint64_t size;
f88e1a42
JS		 3297 char buf[32];
3298
f88e1a42
JS		 3299 bs = bdrv_new("");
3300
96df67d1 3301 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
f88e1a42 3302 if (ret < 0) {
96df67d1 3303 error_report("Could not open '%s'", backing_file->value.s);
f88e1a42
JS		 3304 goto out;
3305 }
3306 bdrv_get_geometry(bs, &size);
3307 size *= 512;
3308
3309 snprintf(buf, sizeof(buf), "%" PRId64, size);
3310 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3311 } else {
3312 error_report("Image creation needs a size parameter");
4f70f249 3313 ret = -EINVAL;
f88e1a42
JS		 3314 goto out;
3315 }
3316 }
3317
3318 printf("Formatting '%s', fmt=%s ", filename, fmt);
3319 print_option_parameters(param);
3320 puts("");
3321
3322 ret = bdrv_create(drv, filename, param);
3323
3324 if (ret < 0) {
3325 if (ret == -ENOTSUP) {
3326 error_report("Formatting or formatting option not supported for "
3327 "file format '%s'", fmt);
3328 } else if (ret == -EFBIG) {
3329 error_report("The image size is too large for file format '%s'",
3330 fmt);
3331 } else {
3332 error_report("%s: error while creating %s: %s", filename, fmt,
3333 strerror(-ret));
3334 }
3335 }
3336
3337out:
3338 free_option_parameters(create_options);
3339 free_option_parameters(param);
3340
3341 if (bs) {
3342 bdrv_delete(bs);
3343 }
4f70f249
JS		 3344
3345 return ret;
f88e1a42 3346}
QEMU mirror