?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'bonzini/scsi-next' into staging
[qemu/opensuse.git] / block.c
blobb88ee9081748a7f37d901ae8d67e99a261d09e6a
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24 #include "config-host.h"
25 #include "qemu-common.h"
26 #include "trace.h"
27 #include "monitor.h"
28 #include "block_int.h"
29 #include "module.h"
30 #include "qjson.h"
31 #include "qemu-coroutine.h"
32 #include "qmp-commands.h"
33 #include "qemu-timer.h"
34
35 #ifdef CONFIG_BSD
36 #include <sys/types.h>
37 #include <sys/stat.h>
38 #include <sys/ioctl.h>
39 #include <sys/queue.h>
40 #ifndef __DragonFly__
41 #include <sys/disk.h>
42 #endif
43 #endif
44
45 #ifdef _WIN32
46 #include <windows.h>
47 #endif
48
49 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
50
51 typedef enum {
52 BDRV_REQ_COPY_ON_READ = 0x1,
53 BDRV_REQ_ZERO_WRITE = 0x2,
54 } BdrvRequestFlags;
55
56 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
57 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
58 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
59 BlockDriverCompletionFunc *cb, void *opaque);
60 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
65 QEMUIOVector *iov);
66 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
68 QEMUIOVector *iov);
69 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
71 BdrvRequestFlags flags);
72 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
76 int64_t sector_num,
77 QEMUIOVector *qiov,
78 int nb_sectors,
79 BlockDriverCompletionFunc *cb,
80 void *opaque,
81 bool is_write);
82 static void coroutine_fn bdrv_co_do_rw(void *opaque);
83
84 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
85 bool is_write, double elapsed_time, uint64_t *wait);
86 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
87 double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
89 bool is_write, int64_t *wait);
90
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92 QTAILQ_HEAD_INITIALIZER(bdrv_states);
93
94 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
95 QLIST_HEAD_INITIALIZER(bdrv_drivers);
96
97 /* The device to use for VM snapshots */
98 static BlockDriverState *bs_snapshots;
99
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist;
102
103 #ifdef _WIN32
104 static int is_windows_drive_prefix(const char *filename)
105 {
106 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
107 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
108 filename[1] == ':');
109 }
110
111 int is_windows_drive(const char *filename)
112 {
113 if (is_windows_drive_prefix(filename) &&
114 filename[2] == '\0')
115 return 1;
116 if (strstart(filename, "\\\\.\\", NULL) ||
117 strstart(filename, "//./", NULL))
118 return 1;
119 return 0;
120 }
121 #endif
122
123 /* throttling disk I/O limits */
124 void bdrv_io_limits_disable(BlockDriverState *bs)
125 {
126 bs->io_limits_enabled = false;
127
128 while (qemu_co_queue_next(&bs->throttled_reqs));
129
130 if (bs->block_timer) {
131 qemu_del_timer(bs->block_timer);
132 qemu_free_timer(bs->block_timer);
133 bs->block_timer = NULL;
134 }
135
136 bs->slice_start = 0;
137 bs->slice_end = 0;
138 bs->slice_time = 0;
139 memset(&bs->io_base, 0, sizeof(bs->io_base));
140 }
141
142 static void bdrv_block_timer(void *opaque)
143 {
144 BlockDriverState *bs = opaque;
145
146 qemu_co_queue_next(&bs->throttled_reqs);
147 }
148
149 void bdrv_io_limits_enable(BlockDriverState *bs)
150 {
151 qemu_co_queue_init(&bs->throttled_reqs);
152 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
153 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
154 bs->slice_start = qemu_get_clock_ns(vm_clock);
155 bs->slice_end = bs->slice_start + bs->slice_time;
156 memset(&bs->io_base, 0, sizeof(bs->io_base));
157 bs->io_limits_enabled = true;
158 }
159
160 bool bdrv_io_limits_enabled(BlockDriverState *bs)
161 {
162 BlockIOLimit *io_limits = &bs->io_limits;
163 return io_limits->bps[BLOCK_IO_LIMIT_READ]
164 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
165 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
166 || io_limits->iops[BLOCK_IO_LIMIT_READ]
167 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
168 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
169 }
170
171 static void bdrv_io_limits_intercept(BlockDriverState *bs,
172 bool is_write, int nb_sectors)
173 {
174 int64_t wait_time = -1;
175
176 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
177 qemu_co_queue_wait(&bs->throttled_reqs);
178 }
179
180 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
181 * throttled requests will not be dequeued until the current request is
182 * allowed to be serviced. So if the current request still exceeds the
183 * limits, it will be inserted to the head. All requests followed it will
184 * be still in throttled_reqs queue.
185 */
186
187 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
188 qemu_mod_timer(bs->block_timer,
189 wait_time + qemu_get_clock_ns(vm_clock));
190 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
191 }
192
193 qemu_co_queue_next(&bs->throttled_reqs);
194 }
195
196 /* check if the path starts with "<protocol>:" */
197 static int path_has_protocol(const char *path)
198 {
199 #ifdef _WIN32
200 if (is_windows_drive(path) ||
201 is_windows_drive_prefix(path)) {
202 return 0;
203 }
204 #endif
205
206 return strchr(path, ':') != NULL;
207 }
208
209 int path_is_absolute(const char *path)
210 {
211 const char *p;
212 #ifdef _WIN32
213 /* specific case for names like: "\\.\d:" */
214 if (*path == '/' || *path == '\\')
215 return 1;
216 #endif
217 p = strchr(path, ':');
218 if (p)
219 p++;
220 else
221 p = path;
222 #ifdef _WIN32
223 return (*p == '/' || *p == '\\');
224 #else
225 return (*p == '/');
226 #endif
227 }
228
229 /* if filename is absolute, just copy it to dest. Otherwise, build a
230 path to it by considering it is relative to base_path. URL are
231 supported. */
232 void path_combine(char *dest, int dest_size,
233 const char *base_path,
234 const char *filename)
235 {
236 const char *p, *p1;
237 int len;
238
239 if (dest_size <= 0)
240 return;
241 if (path_is_absolute(filename)) {
242 pstrcpy(dest, dest_size, filename);
243 } else {
244 p = strchr(base_path, ':');
245 if (p)
246 p++;
247 else
248 p = base_path;
249 p1 = strrchr(base_path, '/');
250 #ifdef _WIN32
251 {
252 const char *p2;
253 p2 = strrchr(base_path, '\\');
254 if (!p1 || p2 > p1)
255 p1 = p2;
256 }
257 #endif
258 if (p1)
259 p1++;
260 else
261 p1 = base_path;
262 if (p1 > p)
263 p = p1;
264 len = p - base_path;
265 if (len > dest_size - 1)
266 len = dest_size - 1;
267 memcpy(dest, base_path, len);
268 dest[len] = '\0';
269 pstrcat(dest, dest_size, filename);
270 }
271 }
272
273 void bdrv_register(BlockDriver *bdrv)
274 {
275 /* Block drivers without coroutine functions need emulation */
276 if (!bdrv->bdrv_co_readv) {
277 bdrv->bdrv_co_readv = bdrv_co_readv_em;
278 bdrv->bdrv_co_writev = bdrv_co_writev_em;
279
280 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
281 * the block driver lacks aio we need to emulate that too.
282 */
283 if (!bdrv->bdrv_aio_readv) {
284 /* add AIO emulation layer */
285 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
286 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
287 }
288 }
289
290 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
291 }
292
293 /* create a new block device (by default it is empty) */
294 BlockDriverState *bdrv_new(const char *device_name)
295 {
296 BlockDriverState *bs;
297
298 bs = g_malloc0(sizeof(BlockDriverState));
299 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
300 if (device_name[0] != '\0') {
301 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
302 }
303 bdrv_iostatus_disable(bs);
304 return bs;
305 }
306
307 BlockDriver *bdrv_find_format(const char *format_name)
308 {
309 BlockDriver *drv1;
310 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
311 if (!strcmp(drv1->format_name, format_name)) {
312 return drv1;
313 }
314 }
315 return NULL;
316 }
317
318 static int bdrv_is_whitelisted(BlockDriver *drv)
319 {
320 static const char *whitelist[] = {
321 CONFIG_BDRV_WHITELIST
322 };
323 const char **p;
324
325 if (!whitelist[0])
326 return 1; /* no whitelist, anything goes */
327
328 for (p = whitelist; *p; p++) {
329 if (!strcmp(drv->format_name, *p)) {
330 return 1;
331 }
332 }
333 return 0;
334 }
335
336 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
337 {
338 BlockDriver *drv = bdrv_find_format(format_name);
339 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
340 }
341
342 int bdrv_create(BlockDriver *drv, const char* filename,
343 QEMUOptionParameter *options)
344 {
345 if (!drv->bdrv_create)
346 return -ENOTSUP;
347
348 return drv->bdrv_create(filename, options);
349 }
350
351 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
352 {
353 BlockDriver *drv;
354
355 drv = bdrv_find_protocol(filename);
356 if (drv == NULL) {
357 return -ENOENT;
358 }
359
360 return bdrv_create(drv, filename, options);
361 }
362
363 #ifdef _WIN32
364 void get_tmp_filename(char *filename, int size)
365 {
366 char temp_dir[MAX_PATH];
367
368 GetTempPath(MAX_PATH, temp_dir);
369 GetTempFileName(temp_dir, "qem", 0, filename);
370 }
371 #else
372 void get_tmp_filename(char *filename, int size)
373 {
374 int fd;
375 const char *tmpdir;
376 /* XXX: race condition possible */
377 tmpdir = getenv("TMPDIR");
378 if (!tmpdir)
379 tmpdir = "/tmp";
380 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
381 fd = mkstemp(filename);
382 close(fd);
383 }
384 #endif
385
386 /*
387 * Detect host devices. By convention, /dev/cdrom[N] is always
388 * recognized as a host CDROM.
389 */
390 static BlockDriver *find_hdev_driver(const char *filename)
391 {
392 int score_max = 0, score;
393 BlockDriver *drv = NULL, *d;
394
395 QLIST_FOREACH(d, &bdrv_drivers, list) {
396 if (d->bdrv_probe_device) {
397 score = d->bdrv_probe_device(filename);
398 if (score > score_max) {
399 score_max = score;
400 drv = d;
401 }
402 }
403 }
404
405 return drv;
406 }
407
408 BlockDriver *bdrv_find_protocol(const char *filename)
409 {
410 BlockDriver *drv1;
411 char protocol[128];
412 int len;
413 const char *p;
414
415 /* TODO Drivers without bdrv_file_open must be specified explicitly */
416
417 /*
418 * XXX(hch): we really should not let host device detection
419 * override an explicit protocol specification, but moving this
420 * later breaks access to device names with colons in them.
421 * Thanks to the brain-dead persistent naming schemes on udev-
422 * based Linux systems those actually are quite common.
423 */
424 drv1 = find_hdev_driver(filename);
425 if (drv1) {
426 return drv1;
427 }
428
429 if (!path_has_protocol(filename)) {
430 return bdrv_find_format("file");
431 }
432 p = strchr(filename, ':');
433 assert(p != NULL);
434 len = p - filename;
435 if (len > sizeof(protocol) - 1)
436 len = sizeof(protocol) - 1;
437 memcpy(protocol, filename, len);
438 protocol[len] = '\0';
439 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
440 if (drv1->protocol_name &&
441 !strcmp(drv1->protocol_name, protocol)) {
442 return drv1;
443 }
444 }
445 return NULL;
446 }
447
448 static int find_image_format(const char *filename, BlockDriver **pdrv)
449 {
450 int ret, score, score_max;
451 BlockDriver *drv1, *drv;
452 uint8_t buf[2048];
453 BlockDriverState *bs;
454
455 ret = bdrv_file_open(&bs, filename, 0);
456 if (ret < 0) {
457 *pdrv = NULL;
458 return ret;
459 }
460
461 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
462 if (bs->sg || !bdrv_is_inserted(bs)) {
463 bdrv_delete(bs);
464 drv = bdrv_find_format("raw");
465 if (!drv) {
466 ret = -ENOENT;
467 }
468 *pdrv = drv;
469 return ret;
470 }
471
472 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
473 bdrv_delete(bs);
474 if (ret < 0) {
475 *pdrv = NULL;
476 return ret;
477 }
478
479 score_max = 0;
480 drv = NULL;
481 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
482 if (drv1->bdrv_probe) {
483 score = drv1->bdrv_probe(buf, ret, filename);
484 if (score > score_max) {
485 score_max = score;
486 drv = drv1;
487 }
488 }
489 }
490 if (!drv) {
491 ret = -ENOENT;
492 }
493 *pdrv = drv;
494 return ret;
495 }
496
497 /**
498 * Set the current 'total_sectors' value
499 */
500 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
501 {
502 BlockDriver *drv = bs->drv;
503
504 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
505 if (bs->sg)
506 return 0;
507
508 /* query actual device if possible, otherwise just trust the hint */
509 if (drv->bdrv_getlength) {
510 int64_t length = drv->bdrv_getlength(bs);
511 if (length < 0) {
512 return length;
513 }
514 hint = length >> BDRV_SECTOR_BITS;
515 }
516
517 bs->total_sectors = hint;
518 return 0;
519 }
520
521 /**
522 * Set open flags for a given cache mode
523 *
524 * Return 0 on success, -1 if the cache mode was invalid.
525 */
526 int bdrv_parse_cache_flags(const char *mode, int *flags)
527 {
528 *flags &= ~BDRV_O_CACHE_MASK;
529
530 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
531 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
532 } else if (!strcmp(mode, "directsync")) {
533 *flags |= BDRV_O_NOCACHE;
534 } else if (!strcmp(mode, "writeback")) {
535 *flags |= BDRV_O_CACHE_WB;
536 } else if (!strcmp(mode, "unsafe")) {
537 *flags |= BDRV_O_CACHE_WB;
538 *flags |= BDRV_O_NO_FLUSH;
539 } else if (!strcmp(mode, "writethrough")) {
540 /* this is the default */
541 } else {
542 return -1;
543 }
544
545 return 0;
546 }
547
548 /**
549 * The copy-on-read flag is actually a reference count so multiple users may
550 * use the feature without worrying about clobbering its previous state.
551 * Copy-on-read stays enabled until all users have called to disable it.
552 */
553 void bdrv_enable_copy_on_read(BlockDriverState *bs)
554 {
555 bs->copy_on_read++;
556 }
557
558 void bdrv_disable_copy_on_read(BlockDriverState *bs)
559 {
560 assert(bs->copy_on_read > 0);
561 bs->copy_on_read--;
562 }
563
564 /*
565 * Common part for opening disk images and files
566 */
567 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
568 int flags, BlockDriver *drv)
569 {
570 int ret, open_flags;
571
572 assert(drv != NULL);
573
574 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
575
576 bs->file = NULL;
577 bs->total_sectors = 0;
578 bs->encrypted = 0;
579 bs->valid_key = 0;
580 bs->sg = 0;
581 bs->open_flags = flags;
582 bs->growable = 0;
583 bs->buffer_alignment = 512;
584
585 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
586 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
587 bdrv_enable_copy_on_read(bs);
588 }
589
590 pstrcpy(bs->filename, sizeof(bs->filename), filename);
591 bs->backing_file[0] = '\0';
592
593 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
594 return -ENOTSUP;
595 }
596
597 bs->drv = drv;
598 bs->opaque = g_malloc0(drv->instance_size);
599
600 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
601
602 /*
603 * Clear flags that are internal to the block layer before opening the
604 * image.
605 */
606 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
607
608 /*
609 * Snapshots should be writable.
610 */
611 if (bs->is_temporary) {
612 open_flags |= BDRV_O_RDWR;
613 }
614
615 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
616
617 /* Open the image, either directly or using a protocol */
618 if (drv->bdrv_file_open) {
619 ret = drv->bdrv_file_open(bs, filename, open_flags);
620 } else {
621 ret = bdrv_file_open(&bs->file, filename, open_flags);
622 if (ret >= 0) {
623 ret = drv->bdrv_open(bs, open_flags);
624 }
625 }
626
627 if (ret < 0) {
628 goto free_and_fail;
629 }
630
631 ret = refresh_total_sectors(bs, bs->total_sectors);
632 if (ret < 0) {
633 goto free_and_fail;
634 }
635
636 #ifndef _WIN32
637 if (bs->is_temporary) {
638 unlink(filename);
639 }
640 #endif
641 return 0;
642
643 free_and_fail:
644 if (bs->file) {
645 bdrv_delete(bs->file);
646 bs->file = NULL;
647 }
648 g_free(bs->opaque);
649 bs->opaque = NULL;
650 bs->drv = NULL;
651 return ret;
652 }
653
654 /*
655 * Opens a file using a protocol (file, host_device, nbd, ...)
656 */
657 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
658 {
659 BlockDriverState *bs;
660 BlockDriver *drv;
661 int ret;
662
663 drv = bdrv_find_protocol(filename);
664 if (!drv) {
665 return -ENOENT;
666 }
667
668 bs = bdrv_new("");
669 ret = bdrv_open_common(bs, filename, flags, drv);
670 if (ret < 0) {
671 bdrv_delete(bs);
672 return ret;
673 }
674 bs->growable = 1;
675 *pbs = bs;
676 return 0;
677 }
678
679 /*
680 * Opens a disk image (raw, qcow2, vmdk, ...)
681 */
682 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
683 BlockDriver *drv)
684 {
685 int ret;
686 char tmp_filename[PATH_MAX];
687
688 if (flags & BDRV_O_SNAPSHOT) {
689 BlockDriverState *bs1;
690 int64_t total_size;
691 int is_protocol = 0;
692 BlockDriver *bdrv_qcow2;
693 QEMUOptionParameter *options;
694 char backing_filename[PATH_MAX];
695
696 /* if snapshot, we create a temporary backing file and open it
697 instead of opening 'filename' directly */
698
699 /* if there is a backing file, use it */
700 bs1 = bdrv_new("");
701 ret = bdrv_open(bs1, filename, 0, drv);
702 if (ret < 0) {
703 bdrv_delete(bs1);
704 return ret;
705 }
706 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
707
708 if (bs1->drv && bs1->drv->protocol_name)
709 is_protocol = 1;
710
711 bdrv_delete(bs1);
712
713 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
714
715 /* Real path is meaningless for protocols */
716 if (is_protocol)
717 snprintf(backing_filename, sizeof(backing_filename),
718 "%s", filename);
719 else if (!realpath(filename, backing_filename))
720 return -errno;
721
722 bdrv_qcow2 = bdrv_find_format("qcow2");
723 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
724
725 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
726 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
727 if (drv) {
728 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
729 drv->format_name);
730 }
731
732 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
733 free_option_parameters(options);
734 if (ret < 0) {
735 return ret;
736 }
737
738 filename = tmp_filename;
739 drv = bdrv_qcow2;
740 bs->is_temporary = 1;
741 }
742
743 /* Find the right image format driver */
744 if (!drv) {
745 ret = find_image_format(filename, &drv);
746 }
747
748 if (!drv) {
749 goto unlink_and_fail;
750 }
751
752 /* Open the image */
753 ret = bdrv_open_common(bs, filename, flags, drv);
754 if (ret < 0) {
755 goto unlink_and_fail;
756 }
757
758 /* If there is a backing file, use it */
759 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
760 char backing_filename[PATH_MAX];
761 int back_flags;
762 BlockDriver *back_drv = NULL;
763
764 bs->backing_hd = bdrv_new("");
765
766 if (path_has_protocol(bs->backing_file)) {
767 pstrcpy(backing_filename, sizeof(backing_filename),
768 bs->backing_file);
769 } else {
770 path_combine(backing_filename, sizeof(backing_filename),
771 filename, bs->backing_file);
772 }
773
774 if (bs->backing_format[0] != '\0') {
775 back_drv = bdrv_find_format(bs->backing_format);
776 }
777
778 /* backing files always opened read-only */
779 back_flags =
780 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
781
782 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
783 if (ret < 0) {
784 bdrv_close(bs);
785 return ret;
786 }
787 if (bs->is_temporary) {
788 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
789 } else {
790 /* base image inherits from "parent" */
791 bs->backing_hd->keep_read_only = bs->keep_read_only;
792 }
793 }
794
795 if (!bdrv_key_required(bs)) {
796 bdrv_dev_change_media_cb(bs, true);
797 }
798
799 /* throttling disk I/O limits */
800 if (bs->io_limits_enabled) {
801 bdrv_io_limits_enable(bs);
802 }
803
804 return 0;
805
806 unlink_and_fail:
807 if (bs->is_temporary) {
808 unlink(filename);
809 }
810 return ret;
811 }
812
813 void bdrv_close(BlockDriverState *bs)
814 {
815 if (bs->drv) {
816 if (bs == bs_snapshots) {
817 bs_snapshots = NULL;
818 }
819 if (bs->backing_hd) {
820 bdrv_delete(bs->backing_hd);
821 bs->backing_hd = NULL;
822 }
823 bs->drv->bdrv_close(bs);
824 g_free(bs->opaque);
825 #ifdef _WIN32
826 if (bs->is_temporary) {
827 unlink(bs->filename);
828 }
829 #endif
830 bs->opaque = NULL;
831 bs->drv = NULL;
832 bs->copy_on_read = 0;
833
834 if (bs->file != NULL) {
835 bdrv_close(bs->file);
836 }
837
838 bdrv_dev_change_media_cb(bs, false);
839 }
840
841 /*throttling disk I/O limits*/
842 if (bs->io_limits_enabled) {
843 bdrv_io_limits_disable(bs);
844 }
845 }
846
847 void bdrv_close_all(void)
848 {
849 BlockDriverState *bs;
850
851 QTAILQ_FOREACH(bs, &bdrv_states, list) {
852 bdrv_close(bs);
853 }
854 }
855
856 /*
857 * Wait for pending requests to complete across all BlockDriverStates
858 *
859 * This function does not flush data to disk, use bdrv_flush_all() for that
860 * after calling this function.
861 */
862 void bdrv_drain_all(void)
863 {
864 BlockDriverState *bs;
865
866 qemu_aio_flush();
867
868 /* If requests are still pending there is a bug somewhere */
869 QTAILQ_FOREACH(bs, &bdrv_states, list) {
870 assert(QLIST_EMPTY(&bs->tracked_requests));
871 assert(qemu_co_queue_empty(&bs->throttled_reqs));
872 }
873 }
874
875 /* make a BlockDriverState anonymous by removing from bdrv_state list.
876 Also, NULL terminate the device_name to prevent double remove */
877 void bdrv_make_anon(BlockDriverState *bs)
878 {
879 if (bs->device_name[0] != '\0') {
880 QTAILQ_REMOVE(&bdrv_states, bs, list);
881 }
882 bs->device_name[0] = '\0';
883 }
884
885 /*
886 * Add new bs contents at the top of an image chain while the chain is
887 * live, while keeping required fields on the top layer.
888 *
889 * This will modify the BlockDriverState fields, and swap contents
890 * between bs_new and bs_top. Both bs_new and bs_top are modified.
891 *
892 * This function does not create any image files.
893 */
894 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
895 {
896 BlockDriverState tmp;
897
898 /* the new bs must not be in bdrv_states */
899 bdrv_make_anon(bs_new);
900
901 tmp = *bs_new;
902
903 /* there are some fields that need to stay on the top layer: */
904
905 /* dev info */
906 tmp.dev_ops = bs_top->dev_ops;
907 tmp.dev_opaque = bs_top->dev_opaque;
908 tmp.dev = bs_top->dev;
909 tmp.buffer_alignment = bs_top->buffer_alignment;
910 tmp.copy_on_read = bs_top->copy_on_read;
911
912 /* i/o timing parameters */
913 tmp.slice_time = bs_top->slice_time;
914 tmp.slice_start = bs_top->slice_start;
915 tmp.slice_end = bs_top->slice_end;
916 tmp.io_limits = bs_top->io_limits;
917 tmp.io_base = bs_top->io_base;
918 tmp.throttled_reqs = bs_top->throttled_reqs;
919 tmp.block_timer = bs_top->block_timer;
920 tmp.io_limits_enabled = bs_top->io_limits_enabled;
921
922 /* geometry */
923 tmp.cyls = bs_top->cyls;
924 tmp.heads = bs_top->heads;
925 tmp.secs = bs_top->secs;
926 tmp.translation = bs_top->translation;
927
928 /* r/w error */
929 tmp.on_read_error = bs_top->on_read_error;
930 tmp.on_write_error = bs_top->on_write_error;
931
932 /* i/o status */
933 tmp.iostatus_enabled = bs_top->iostatus_enabled;
934 tmp.iostatus = bs_top->iostatus;
935
936 /* keep the same entry in bdrv_states */
937 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
938 tmp.list = bs_top->list;
939
940 /* The contents of 'tmp' will become bs_top, as we are
941 * swapping bs_new and bs_top contents. */
942 tmp.backing_hd = bs_new;
943 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
944
945 /* swap contents of the fixed new bs and the current top */
946 *bs_new = *bs_top;
947 *bs_top = tmp;
948
949 /* clear the copied fields in the new backing file */
950 bdrv_detach_dev(bs_new, bs_new->dev);
951
952 qemu_co_queue_init(&bs_new->throttled_reqs);
953 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
954 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
955 bdrv_iostatus_disable(bs_new);
956
957 /* we don't use bdrv_io_limits_disable() for this, because we don't want
958 * to affect or delete the block_timer, as it has been moved to bs_top */
959 bs_new->io_limits_enabled = false;
960 bs_new->block_timer = NULL;
961 bs_new->slice_time = 0;
962 bs_new->slice_start = 0;
963 bs_new->slice_end = 0;
964 }
965
966 void bdrv_delete(BlockDriverState *bs)
967 {
968 assert(!bs->dev);
969
970 /* remove from list, if necessary */
971 bdrv_make_anon(bs);
972
973 bdrv_close(bs);
974 if (bs->file != NULL) {
975 bdrv_delete(bs->file);
976 }
977
978 assert(bs != bs_snapshots);
979 g_free(bs);
980 }
981
982 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
983 /* TODO change to DeviceState *dev when all users are qdevified */
984 {
985 if (bs->dev) {
986 return -EBUSY;
987 }
988 bs->dev = dev;
989 bdrv_iostatus_reset(bs);
990 return 0;
991 }
992
993 /* TODO qdevified devices don't use this, remove when devices are qdevified */
994 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
995 {
996 if (bdrv_attach_dev(bs, dev) < 0) {
997 abort();
998 }
999 }
1000
1001 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1002 /* TODO change to DeviceState *dev when all users are qdevified */
1003 {
1004 assert(bs->dev == dev);
1005 bs->dev = NULL;
1006 bs->dev_ops = NULL;
1007 bs->dev_opaque = NULL;
1008 bs->buffer_alignment = 512;
1009 }
1010
1011 /* TODO change to return DeviceState * when all users are qdevified */
1012 void *bdrv_get_attached_dev(BlockDriverState *bs)
1013 {
1014 return bs->dev;
1015 }
1016
1017 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1018 void *opaque)
1019 {
1020 bs->dev_ops = ops;
1021 bs->dev_opaque = opaque;
1022 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1023 bs_snapshots = NULL;
1024 }
1025 }
1026
1027 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1028 BlockQMPEventAction action, int is_read)
1029 {
1030 QObject *data;
1031 const char *action_str;
1032
1033 switch (action) {
1034 case BDRV_ACTION_REPORT:
1035 action_str = "report";
1036 break;
1037 case BDRV_ACTION_IGNORE:
1038 action_str = "ignore";
1039 break;
1040 case BDRV_ACTION_STOP:
1041 action_str = "stop";
1042 break;
1043 default:
1044 abort();
1045 }
1046
1047 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1048 bdrv->device_name,
1049 action_str,
1050 is_read ? "read" : "write");
1051 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1052
1053 qobject_decref(data);
1054 }
1055
1056 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1057 {
1058 QObject *data;
1059
1060 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1061 bdrv_get_device_name(bs), ejected);
1062 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1063
1064 qobject_decref(data);
1065 }
1066
1067 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1068 {
1069 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1070 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1071 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1072 if (tray_was_closed) {
1073 /* tray open */
1074 bdrv_emit_qmp_eject_event(bs, true);
1075 }
1076 if (load) {
1077 /* tray close */
1078 bdrv_emit_qmp_eject_event(bs, false);
1079 }
1080 }
1081 }
1082
1083 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1084 {
1085 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1086 }
1087
1088 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1089 {
1090 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1091 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1092 }
1093 }
1094
1095 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1096 {
1097 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1098 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1099 }
1100 return false;
1101 }
1102
1103 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1104 {
1105 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1106 bs->dev_ops->resize_cb(bs->dev_opaque);
1107 }
1108 }
1109
1110 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1111 {
1112 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1113 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1114 }
1115 return false;
1116 }
1117
1118 /*
1119 * Run consistency checks on an image
1120 *
1121 * Returns 0 if the check could be completed (it doesn't mean that the image is
1122 * free of errors) or -errno when an internal error occurred. The results of the
1123 * check are stored in res.
1124 */
1125 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1126 {
1127 if (bs->drv->bdrv_check == NULL) {
1128 return -ENOTSUP;
1129 }
1130
1131 memset(res, 0, sizeof(*res));
1132 return bs->drv->bdrv_check(bs, res);
1133 }
1134
1135 #define COMMIT_BUF_SECTORS 2048
1136
1137 /* commit COW file into the raw image */
1138 int bdrv_commit(BlockDriverState *bs)
1139 {
1140 BlockDriver *drv = bs->drv;
1141 BlockDriver *backing_drv;
1142 int64_t sector, total_sectors;
1143 int n, ro, open_flags;
1144 int ret = 0, rw_ret = 0;
1145 uint8_t *buf;
1146 char filename[1024];
1147 BlockDriverState *bs_rw, *bs_ro;
1148
1149 if (!drv)
1150 return -ENOMEDIUM;
1151
1152 if (!bs->backing_hd) {
1153 return -ENOTSUP;
1154 }
1155
1156 if (bs->backing_hd->keep_read_only) {
1157 return -EACCES;
1158 }
1159
1160 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1161 return -EBUSY;
1162 }
1163
1164 backing_drv = bs->backing_hd->drv;
1165 ro = bs->backing_hd->read_only;
1166 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1167 open_flags = bs->backing_hd->open_flags;
1168
1169 if (ro) {
1170 /* re-open as RW */
1171 bdrv_delete(bs->backing_hd);
1172 bs->backing_hd = NULL;
1173 bs_rw = bdrv_new("");
1174 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1175 backing_drv);
1176 if (rw_ret < 0) {
1177 bdrv_delete(bs_rw);
1178 /* try to re-open read-only */
1179 bs_ro = bdrv_new("");
1180 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1181 backing_drv);
1182 if (ret < 0) {
1183 bdrv_delete(bs_ro);
1184 /* drive not functional anymore */
1185 bs->drv = NULL;
1186 return ret;
1187 }
1188 bs->backing_hd = bs_ro;
1189 return rw_ret;
1190 }
1191 bs->backing_hd = bs_rw;
1192 }
1193
1194 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1195 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1196
1197 for (sector = 0; sector < total_sectors; sector += n) {
1198 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1199
1200 if (bdrv_read(bs, sector, buf, n) != 0) {
1201 ret = -EIO;
1202 goto ro_cleanup;
1203 }
1204
1205 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1206 ret = -EIO;
1207 goto ro_cleanup;
1208 }
1209 }
1210 }
1211
1212 if (drv->bdrv_make_empty) {
1213 ret = drv->bdrv_make_empty(bs);
1214 bdrv_flush(bs);
1215 }
1216
1217 /*
1218 * Make sure all data we wrote to the backing device is actually
1219 * stable on disk.
1220 */
1221 if (bs->backing_hd)
1222 bdrv_flush(bs->backing_hd);
1223
1224 ro_cleanup:
1225 g_free(buf);
1226
1227 if (ro) {
1228 /* re-open as RO */
1229 bdrv_delete(bs->backing_hd);
1230 bs->backing_hd = NULL;
1231 bs_ro = bdrv_new("");
1232 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1233 backing_drv);
1234 if (ret < 0) {
1235 bdrv_delete(bs_ro);
1236 /* drive not functional anymore */
1237 bs->drv = NULL;
1238 return ret;
1239 }
1240 bs->backing_hd = bs_ro;
1241 bs->backing_hd->keep_read_only = 0;
1242 }
1243
1244 return ret;
1245 }
1246
1247 int bdrv_commit_all(void)
1248 {
1249 BlockDriverState *bs;
1250
1251 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1252 int ret = bdrv_commit(bs);
1253 if (ret < 0) {
1254 return ret;
1255 }
1256 }
1257 return 0;
1258 }
1259
1260 struct BdrvTrackedRequest {
1261 BlockDriverState *bs;
1262 int64_t sector_num;
1263 int nb_sectors;
1264 bool is_write;
1265 QLIST_ENTRY(BdrvTrackedRequest) list;
1266 Coroutine *co; /* owner, used for deadlock detection */
1267 CoQueue wait_queue; /* coroutines blocked on this request */
1268 };
1269
1270 /**
1271 * Remove an active request from the tracked requests list
1272 *
1273 * This function should be called when a tracked request is completing.
1274 */
1275 static void tracked_request_end(BdrvTrackedRequest *req)
1276 {
1277 QLIST_REMOVE(req, list);
1278 qemu_co_queue_restart_all(&req->wait_queue);
1279 }
1280
1281 /**
1282 * Add an active request to the tracked requests list
1283 */
1284 static void tracked_request_begin(BdrvTrackedRequest *req,
1285 BlockDriverState *bs,
1286 int64_t sector_num,
1287 int nb_sectors, bool is_write)
1288 {
1289 *req = (BdrvTrackedRequest){
1290 .bs = bs,
1291 .sector_num = sector_num,
1292 .nb_sectors = nb_sectors,
1293 .is_write = is_write,
1294 .co = qemu_coroutine_self(),
1295 };
1296
1297 qemu_co_queue_init(&req->wait_queue);
1298
1299 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1300 }
1301
1302 /**
1303 * Round a region to cluster boundaries
1304 */
1305 static void round_to_clusters(BlockDriverState *bs,
1306 int64_t sector_num, int nb_sectors,
1307 int64_t *cluster_sector_num,
1308 int *cluster_nb_sectors)
1309 {
1310 BlockDriverInfo bdi;
1311
1312 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1313 *cluster_sector_num = sector_num;
1314 *cluster_nb_sectors = nb_sectors;
1315 } else {
1316 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1317 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1318 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1319 nb_sectors, c);
1320 }
1321 }
1322
1323 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1324 int64_t sector_num, int nb_sectors) {
1325 /* aaaa bbbb */
1326 if (sector_num >= req->sector_num + req->nb_sectors) {
1327 return false;
1328 }
1329 /* bbbb aaaa */
1330 if (req->sector_num >= sector_num + nb_sectors) {
1331 return false;
1332 }
1333 return true;
1334 }
1335
1336 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1337 int64_t sector_num, int nb_sectors)
1338 {
1339 BdrvTrackedRequest *req;
1340 int64_t cluster_sector_num;
1341 int cluster_nb_sectors;
1342 bool retry;
1343
1344 /* If we touch the same cluster it counts as an overlap. This guarantees
1345 * that allocating writes will be serialized and not race with each other
1346 * for the same cluster. For example, in copy-on-read it ensures that the
1347 * CoR read and write operations are atomic and guest writes cannot
1348 * interleave between them.
1349 */
1350 round_to_clusters(bs, sector_num, nb_sectors,
1351 &cluster_sector_num, &cluster_nb_sectors);
1352
1353 do {
1354 retry = false;
1355 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1356 if (tracked_request_overlaps(req, cluster_sector_num,
1357 cluster_nb_sectors)) {
1358 /* Hitting this means there was a reentrant request, for
1359 * example, a block driver issuing nested requests. This must
1360 * never happen since it means deadlock.
1361 */
1362 assert(qemu_coroutine_self() != req->co);
1363
1364 qemu_co_queue_wait(&req->wait_queue);
1365 retry = true;
1366 break;
1367 }
1368 }
1369 } while (retry);
1370 }
1371
1372 /*
1373 * Return values:
1374 * 0 - success
1375 * -EINVAL - backing format specified, but no file
1376 * -ENOSPC - can't update the backing file because no space is left in the
1377 * image file header
1378 * -ENOTSUP - format driver doesn't support changing the backing file
1379 */
1380 int bdrv_change_backing_file(BlockDriverState *bs,
1381 const char *backing_file, const char *backing_fmt)
1382 {
1383 BlockDriver *drv = bs->drv;
1384
1385 if (drv->bdrv_change_backing_file != NULL) {
1386 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1387 } else {
1388 return -ENOTSUP;
1389 }
1390 }
1391
1392 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1393 size_t size)
1394 {
1395 int64_t len;
1396
1397 if (!bdrv_is_inserted(bs))
1398 return -ENOMEDIUM;
1399
1400 if (bs->growable)
1401 return 0;
1402
1403 len = bdrv_getlength(bs);
1404
1405 if (offset < 0)
1406 return -EIO;
1407
1408 if ((offset > len) || (len - offset < size))
1409 return -EIO;
1410
1411 return 0;
1412 }
1413
1414 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1415 int nb_sectors)
1416 {
1417 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1418 nb_sectors * BDRV_SECTOR_SIZE);
1419 }
1420
1421 typedef struct RwCo {
1422 BlockDriverState *bs;
1423 int64_t sector_num;
1424 int nb_sectors;
1425 QEMUIOVector *qiov;
1426 bool is_write;
1427 int ret;
1428 } RwCo;
1429
1430 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1431 {
1432 RwCo *rwco = opaque;
1433
1434 if (!rwco->is_write) {
1435 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1436 rwco->nb_sectors, rwco->qiov, 0);
1437 } else {
1438 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1439 rwco->nb_sectors, rwco->qiov, 0);
1440 }
1441 }
1442
1443 /*
1444 * Process a synchronous request using coroutines
1445 */
1446 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1447 int nb_sectors, bool is_write)
1448 {
1449 QEMUIOVector qiov;
1450 struct iovec iov = {
1451 .iov_base = (void *)buf,
1452 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1453 };
1454 Coroutine *co;
1455 RwCo rwco = {
1456 .bs = bs,
1457 .sector_num = sector_num,
1458 .nb_sectors = nb_sectors,
1459 .qiov = &qiov,
1460 .is_write = is_write,
1461 .ret = NOT_DONE,
1462 };
1463
1464 qemu_iovec_init_external(&qiov, &iov, 1);
1465
1466 if (qemu_in_coroutine()) {
1467 /* Fast-path if already in coroutine context */
1468 bdrv_rw_co_entry(&rwco);
1469 } else {
1470 co = qemu_coroutine_create(bdrv_rw_co_entry);
1471 qemu_coroutine_enter(co, &rwco);
1472 while (rwco.ret == NOT_DONE) {
1473 qemu_aio_wait();
1474 }
1475 }
1476 return rwco.ret;
1477 }
1478
1479 /* return < 0 if error. See bdrv_write() for the return codes */
1480 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1481 uint8_t *buf, int nb_sectors)
1482 {
1483 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1484 }
1485
1486 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1487 int nb_sectors, int dirty)
1488 {
1489 int64_t start, end;
1490 unsigned long val, idx, bit;
1491
1492 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1493 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1494
1495 for (; start <= end; start++) {
1496 idx = start / (sizeof(unsigned long) * 8);
1497 bit = start % (sizeof(unsigned long) * 8);
1498 val = bs->dirty_bitmap[idx];
1499 if (dirty) {
1500 if (!(val & (1UL << bit))) {
1501 bs->dirty_count++;
1502 val |= 1UL << bit;
1503 }
1504 } else {
1505 if (val & (1UL << bit)) {
1506 bs->dirty_count--;
1507 val &= ~(1UL << bit);
1508 }
1509 }
1510 bs->dirty_bitmap[idx] = val;
1511 }
1512 }
1513
1514 /* Return < 0 if error. Important errors are:
1515 -EIO generic I/O error (may happen for all errors)
1516 -ENOMEDIUM No media inserted.
1517 -EINVAL Invalid sector number or nb_sectors
1518 -EACCES Trying to write a read-only device
1519 */
1520 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1521 const uint8_t *buf, int nb_sectors)
1522 {
1523 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1524 }
1525
1526 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1527 void *buf, int count1)
1528 {
1529 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1530 int len, nb_sectors, count;
1531 int64_t sector_num;
1532 int ret;
1533
1534 count = count1;
1535 /* first read to align to sector start */
1536 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1537 if (len > count)
1538 len = count;
1539 sector_num = offset >> BDRV_SECTOR_BITS;
1540 if (len > 0) {
1541 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1542 return ret;
1543 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1544 count -= len;
1545 if (count == 0)
1546 return count1;
1547 sector_num++;
1548 buf += len;
1549 }
1550
1551 /* read the sectors "in place" */
1552 nb_sectors = count >> BDRV_SECTOR_BITS;
1553 if (nb_sectors > 0) {
1554 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1555 return ret;
1556 sector_num += nb_sectors;
1557 len = nb_sectors << BDRV_SECTOR_BITS;
1558 buf += len;
1559 count -= len;
1560 }
1561
1562 /* add data from the last sector */
1563 if (count > 0) {
1564 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1565 return ret;
1566 memcpy(buf, tmp_buf, count);
1567 }
1568 return count1;
1569 }
1570
1571 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1572 const void *buf, int count1)
1573 {
1574 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1575 int len, nb_sectors, count;
1576 int64_t sector_num;
1577 int ret;
1578
1579 count = count1;
1580 /* first write to align to sector start */
1581 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1582 if (len > count)
1583 len = count;
1584 sector_num = offset >> BDRV_SECTOR_BITS;
1585 if (len > 0) {
1586 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1587 return ret;
1588 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1589 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1590 return ret;
1591 count -= len;
1592 if (count == 0)
1593 return count1;
1594 sector_num++;
1595 buf += len;
1596 }
1597
1598 /* write the sectors "in place" */
1599 nb_sectors = count >> BDRV_SECTOR_BITS;
1600 if (nb_sectors > 0) {
1601 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1602 return ret;
1603 sector_num += nb_sectors;
1604 len = nb_sectors << BDRV_SECTOR_BITS;
1605 buf += len;
1606 count -= len;
1607 }
1608
1609 /* add data from the last sector */
1610 if (count > 0) {
1611 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1612 return ret;
1613 memcpy(tmp_buf, buf, count);
1614 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1615 return ret;
1616 }
1617 return count1;
1618 }
1619
1620 /*
1621 * Writes to the file and ensures that no writes are reordered across this
1622 * request (acts as a barrier)
1623 *
1624 * Returns 0 on success, -errno in error cases.
1625 */
1626 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1627 const void *buf, int count)
1628 {
1629 int ret;
1630
1631 ret = bdrv_pwrite(bs, offset, buf, count);
1632 if (ret < 0) {
1633 return ret;
1634 }
1635
1636 /* No flush needed for cache modes that use O_DSYNC */
1637 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1638 bdrv_flush(bs);
1639 }
1640
1641 return 0;
1642 }
1643
1644 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1645 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1646 {
1647 /* Perform I/O through a temporary buffer so that users who scribble over
1648 * their read buffer while the operation is in progress do not end up
1649 * modifying the image file. This is critical for zero-copy guest I/O
1650 * where anything might happen inside guest memory.
1651 */
1652 void *bounce_buffer;
1653
1654 BlockDriver *drv = bs->drv;
1655 struct iovec iov;
1656 QEMUIOVector bounce_qiov;
1657 int64_t cluster_sector_num;
1658 int cluster_nb_sectors;
1659 size_t skip_bytes;
1660 int ret;
1661
1662 /* Cover entire cluster so no additional backing file I/O is required when
1663 * allocating cluster in the image file.
1664 */
1665 round_to_clusters(bs, sector_num, nb_sectors,
1666 &cluster_sector_num, &cluster_nb_sectors);
1667
1668 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1669 cluster_sector_num, cluster_nb_sectors);
1670
1671 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1672 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1673 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1674
1675 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1676 &bounce_qiov);
1677 if (ret < 0) {
1678 goto err;
1679 }
1680
1681 if (drv->bdrv_co_write_zeroes &&
1682 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1683 ret = drv->bdrv_co_write_zeroes(bs, cluster_sector_num,
1684 cluster_nb_sectors);
1685 } else {
1686 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1687 &bounce_qiov);
1688 }
1689
1690 if (ret < 0) {
1691 /* It might be okay to ignore write errors for guest requests. If this
1692 * is a deliberate copy-on-read then we don't want to ignore the error.
1693 * Simply report it in all cases.
1694 */
1695 goto err;
1696 }
1697
1698 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1699 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1700 nb_sectors * BDRV_SECTOR_SIZE);
1701
1702 err:
1703 qemu_vfree(bounce_buffer);
1704 return ret;
1705 }
1706
1707 /*
1708 * Handle a read request in coroutine context
1709 */
1710 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1711 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1712 BdrvRequestFlags flags)
1713 {
1714 BlockDriver *drv = bs->drv;
1715 BdrvTrackedRequest req;
1716 int ret;
1717
1718 if (!drv) {
1719 return -ENOMEDIUM;
1720 }
1721 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1722 return -EIO;
1723 }
1724
1725 /* throttling disk read I/O */
1726 if (bs->io_limits_enabled) {
1727 bdrv_io_limits_intercept(bs, false, nb_sectors);
1728 }
1729
1730 if (bs->copy_on_read) {
1731 flags |= BDRV_REQ_COPY_ON_READ;
1732 }
1733 if (flags & BDRV_REQ_COPY_ON_READ) {
1734 bs->copy_on_read_in_flight++;
1735 }
1736
1737 if (bs->copy_on_read_in_flight) {
1738 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1739 }
1740
1741 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1742
1743 if (flags & BDRV_REQ_COPY_ON_READ) {
1744 int pnum;
1745
1746 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1747 if (ret < 0) {
1748 goto out;
1749 }
1750
1751 if (!ret || pnum != nb_sectors) {
1752 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1753 goto out;
1754 }
1755 }
1756
1757 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1758
1759 out:
1760 tracked_request_end(&req);
1761
1762 if (flags & BDRV_REQ_COPY_ON_READ) {
1763 bs->copy_on_read_in_flight--;
1764 }
1765
1766 return ret;
1767 }
1768
1769 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1770 int nb_sectors, QEMUIOVector *qiov)
1771 {
1772 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1773
1774 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1775 }
1776
1777 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1778 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1779 {
1780 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1781
1782 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1783 BDRV_REQ_COPY_ON_READ);
1784 }
1785
1786 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1787 int64_t sector_num, int nb_sectors)
1788 {
1789 BlockDriver *drv = bs->drv;
1790 QEMUIOVector qiov;
1791 struct iovec iov;
1792 int ret;
1793
1794 /* First try the efficient write zeroes operation */
1795 if (drv->bdrv_co_write_zeroes) {
1796 return drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1797 }
1798
1799 /* Fall back to bounce buffer if write zeroes is unsupported */
1800 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1801 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1802 memset(iov.iov_base, 0, iov.iov_len);
1803 qemu_iovec_init_external(&qiov, &iov, 1);
1804
1805 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1806
1807 qemu_vfree(iov.iov_base);
1808 return ret;
1809 }
1810
1811 /*
1812 * Handle a write request in coroutine context
1813 */
1814 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1815 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1816 BdrvRequestFlags flags)
1817 {
1818 BlockDriver *drv = bs->drv;
1819 BdrvTrackedRequest req;
1820 int ret;
1821
1822 if (!bs->drv) {
1823 return -ENOMEDIUM;
1824 }
1825 if (bs->read_only) {
1826 return -EACCES;
1827 }
1828 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1829 return -EIO;
1830 }
1831
1832 /* throttling disk write I/O */
1833 if (bs->io_limits_enabled) {
1834 bdrv_io_limits_intercept(bs, true, nb_sectors);
1835 }
1836
1837 if (bs->copy_on_read_in_flight) {
1838 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1839 }
1840
1841 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1842
1843 if (flags & BDRV_REQ_ZERO_WRITE) {
1844 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1845 } else {
1846 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1847 }
1848
1849 if (bs->dirty_bitmap) {
1850 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1851 }
1852
1853 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1854 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1855 }
1856
1857 tracked_request_end(&req);
1858
1859 return ret;
1860 }
1861
1862 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1863 int nb_sectors, QEMUIOVector *qiov)
1864 {
1865 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1866
1867 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1868 }
1869
1870 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1871 int64_t sector_num, int nb_sectors)
1872 {
1873 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1874
1875 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1876 BDRV_REQ_ZERO_WRITE);
1877 }
1878
1879 /**
1880 * Truncate file to 'offset' bytes (needed only for file protocols)
1881 */
1882 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1883 {
1884 BlockDriver *drv = bs->drv;
1885 int ret;
1886 if (!drv)
1887 return -ENOMEDIUM;
1888 if (!drv->bdrv_truncate)
1889 return -ENOTSUP;
1890 if (bs->read_only)
1891 return -EACCES;
1892 if (bdrv_in_use(bs))
1893 return -EBUSY;
1894 ret = drv->bdrv_truncate(bs, offset);
1895 if (ret == 0) {
1896 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1897 bdrv_dev_resize_cb(bs);
1898 }
1899 return ret;
1900 }
1901
1902 /**
1903 * Length of a allocated file in bytes. Sparse files are counted by actual
1904 * allocated space. Return < 0 if error or unknown.
1905 */
1906 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1907 {
1908 BlockDriver *drv = bs->drv;
1909 if (!drv) {
1910 return -ENOMEDIUM;
1911 }
1912 if (drv->bdrv_get_allocated_file_size) {
1913 return drv->bdrv_get_allocated_file_size(bs);
1914 }
1915 if (bs->file) {
1916 return bdrv_get_allocated_file_size(bs->file);
1917 }
1918 return -ENOTSUP;
1919 }
1920
1921 /**
1922 * Length of a file in bytes. Return < 0 if error or unknown.
1923 */
1924 int64_t bdrv_getlength(BlockDriverState *bs)
1925 {
1926 BlockDriver *drv = bs->drv;
1927 if (!drv)
1928 return -ENOMEDIUM;
1929
1930 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
1931 if (drv->bdrv_getlength) {
1932 return drv->bdrv_getlength(bs);
1933 }
1934 }
1935 return bs->total_sectors * BDRV_SECTOR_SIZE;
1936 }
1937
1938 /* return 0 as number of sectors if no device present or error */
1939 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1940 {
1941 int64_t length;
1942 length = bdrv_getlength(bs);
1943 if (length < 0)
1944 length = 0;
1945 else
1946 length = length >> BDRV_SECTOR_BITS;
1947 *nb_sectors_ptr = length;
1948 }
1949
1950 struct partition {
1951 uint8_t boot_ind; /* 0x80 - active */
1952 uint8_t head; /* starting head */
1953 uint8_t sector; /* starting sector */
1954 uint8_t cyl; /* starting cylinder */
1955 uint8_t sys_ind; /* What partition type */
1956 uint8_t end_head; /* end head */
1957 uint8_t end_sector; /* end sector */
1958 uint8_t end_cyl; /* end cylinder */
1959 uint32_t start_sect; /* starting sector counting from 0 */
1960 uint32_t nr_sects; /* nr of sectors in partition */
1961 } QEMU_PACKED;
1962
1963 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1964 static int guess_disk_lchs(BlockDriverState *bs,
1965 int *pcylinders, int *pheads, int *psectors)
1966 {
1967 uint8_t buf[BDRV_SECTOR_SIZE];
1968 int ret, i, heads, sectors, cylinders;
1969 struct partition *p;
1970 uint32_t nr_sects;
1971 uint64_t nb_sectors;
1972
1973 bdrv_get_geometry(bs, &nb_sectors);
1974
1975 ret = bdrv_read(bs, 0, buf, 1);
1976 if (ret < 0)
1977 return -1;
1978 /* test msdos magic */
1979 if (buf[510] != 0x55 || buf[511] != 0xaa)
1980 return -1;
1981 for(i = 0; i < 4; i++) {
1982 p = ((struct partition *)(buf + 0x1be)) + i;
1983 nr_sects = le32_to_cpu(p->nr_sects);
1984 if (nr_sects && p->end_head) {
1985 /* We make the assumption that the partition terminates on
1986 a cylinder boundary */
1987 heads = p->end_head + 1;
1988 sectors = p->end_sector & 63;
1989 if (sectors == 0)
1990 continue;
1991 cylinders = nb_sectors / (heads * sectors);
1992 if (cylinders < 1 || cylinders > 16383)
1993 continue;
1994 *pheads = heads;
1995 *psectors = sectors;
1996 *pcylinders = cylinders;
1997 #if 0
1998 printf("guessed geometry: LCHS=%d %d %d\n",
1999 cylinders, heads, sectors);
2000 #endif
2001 return 0;
2002 }
2003 }
2004 return -1;
2005 }
2006
2007 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2008 {
2009 int translation, lba_detected = 0;
2010 int cylinders, heads, secs;
2011 uint64_t nb_sectors;
2012
2013 /* if a geometry hint is available, use it */
2014 bdrv_get_geometry(bs, &nb_sectors);
2015 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2016 translation = bdrv_get_translation_hint(bs);
2017 if (cylinders != 0) {
2018 *pcyls = cylinders;
2019 *pheads = heads;
2020 *psecs = secs;
2021 } else {
2022 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2023 if (heads > 16) {
2024 /* if heads > 16, it means that a BIOS LBA
2025 translation was active, so the default
2026 hardware geometry is OK */
2027 lba_detected = 1;
2028 goto default_geometry;
2029 } else {
2030 *pcyls = cylinders;
2031 *pheads = heads;
2032 *psecs = secs;
2033 /* disable any translation to be in sync with
2034 the logical geometry */
2035 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2036 bdrv_set_translation_hint(bs,
2037 BIOS_ATA_TRANSLATION_NONE);
2038 }
2039 }
2040 } else {
2041 default_geometry:
2042 /* if no geometry, use a standard physical disk geometry */
2043 cylinders = nb_sectors / (16 * 63);
2044
2045 if (cylinders > 16383)
2046 cylinders = 16383;
2047 else if (cylinders < 2)
2048 cylinders = 2;
2049 *pcyls = cylinders;
2050 *pheads = 16;
2051 *psecs = 63;
2052 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2053 if ((*pcyls * *pheads) <= 131072) {
2054 bdrv_set_translation_hint(bs,
2055 BIOS_ATA_TRANSLATION_LARGE);
2056 } else {
2057 bdrv_set_translation_hint(bs,
2058 BIOS_ATA_TRANSLATION_LBA);
2059 }
2060 }
2061 }
2062 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2063 }
2064 }
2065
2066 void bdrv_set_geometry_hint(BlockDriverState *bs,
2067 int cyls, int heads, int secs)
2068 {
2069 bs->cyls = cyls;
2070 bs->heads = heads;
2071 bs->secs = secs;
2072 }
2073
2074 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2075 {
2076 bs->translation = translation;
2077 }
2078
2079 void bdrv_get_geometry_hint(BlockDriverState *bs,
2080 int *pcyls, int *pheads, int *psecs)
2081 {
2082 *pcyls = bs->cyls;
2083 *pheads = bs->heads;
2084 *psecs = bs->secs;
2085 }
2086
2087 /* throttling disk io limits */
2088 void bdrv_set_io_limits(BlockDriverState *bs,
2089 BlockIOLimit *io_limits)
2090 {
2091 bs->io_limits = *io_limits;
2092 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2093 }
2094
2095 /* Recognize floppy formats */
2096 typedef struct FDFormat {
2097 FDriveType drive;
2098 uint8_t last_sect;
2099 uint8_t max_track;
2100 uint8_t max_head;
2101 FDriveRate rate;
2102 } FDFormat;
2103
2104 static const FDFormat fd_formats[] = {
2105 /* First entry is default format */
2106 /* 1.44 MB 3"1/2 floppy disks */
2107 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2108 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2109 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2110 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2111 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2112 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2113 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2114 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2115 /* 2.88 MB 3"1/2 floppy disks */
2116 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2117 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2118 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2119 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2120 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2121 /* 720 kB 3"1/2 floppy disks */
2122 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2123 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2124 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2125 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2126 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2127 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2128 /* 1.2 MB 5"1/4 floppy disks */
2129 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2130 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2131 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2132 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2133 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2134 /* 720 kB 5"1/4 floppy disks */
2135 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2136 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2137 /* 360 kB 5"1/4 floppy disks */
2138 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2139 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2140 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2141 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2142 /* 320 kB 5"1/4 floppy disks */
2143 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2144 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2145 /* 360 kB must match 5"1/4 better than 3"1/2... */
2146 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2147 /* end */
2148 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2149 };
2150
2151 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2152 int *max_track, int *last_sect,
2153 FDriveType drive_in, FDriveType *drive,
2154 FDriveRate *rate)
2155 {
2156 const FDFormat *parse;
2157 uint64_t nb_sectors, size;
2158 int i, first_match, match;
2159
2160 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2161 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2162 /* User defined disk */
2163 *rate = FDRIVE_RATE_500K;
2164 } else {
2165 bdrv_get_geometry(bs, &nb_sectors);
2166 match = -1;
2167 first_match = -1;
2168 for (i = 0; ; i++) {
2169 parse = &fd_formats[i];
2170 if (parse->drive == FDRIVE_DRV_NONE) {
2171 break;
2172 }
2173 if (drive_in == parse->drive ||
2174 drive_in == FDRIVE_DRV_NONE) {
2175 size = (parse->max_head + 1) * parse->max_track *
2176 parse->last_sect;
2177 if (nb_sectors == size) {
2178 match = i;
2179 break;
2180 }
2181 if (first_match == -1) {
2182 first_match = i;
2183 }
2184 }
2185 }
2186 if (match == -1) {
2187 if (first_match == -1) {
2188 match = 1;
2189 } else {
2190 match = first_match;
2191 }
2192 parse = &fd_formats[match];
2193 }
2194 *nb_heads = parse->max_head + 1;
2195 *max_track = parse->max_track;
2196 *last_sect = parse->last_sect;
2197 *drive = parse->drive;
2198 *rate = parse->rate;
2199 }
2200 }
2201
2202 int bdrv_get_translation_hint(BlockDriverState *bs)
2203 {
2204 return bs->translation;
2205 }
2206
2207 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2208 BlockErrorAction on_write_error)
2209 {
2210 bs->on_read_error = on_read_error;
2211 bs->on_write_error = on_write_error;
2212 }
2213
2214 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2215 {
2216 return is_read ? bs->on_read_error : bs->on_write_error;
2217 }
2218
2219 int bdrv_is_read_only(BlockDriverState *bs)
2220 {
2221 return bs->read_only;
2222 }
2223
2224 int bdrv_is_sg(BlockDriverState *bs)
2225 {
2226 return bs->sg;
2227 }
2228
2229 int bdrv_enable_write_cache(BlockDriverState *bs)
2230 {
2231 return bs->enable_write_cache;
2232 }
2233
2234 int bdrv_is_encrypted(BlockDriverState *bs)
2235 {
2236 if (bs->backing_hd && bs->backing_hd->encrypted)
2237 return 1;
2238 return bs->encrypted;
2239 }
2240
2241 int bdrv_key_required(BlockDriverState *bs)
2242 {
2243 BlockDriverState *backing_hd = bs->backing_hd;
2244
2245 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2246 return 1;
2247 return (bs->encrypted && !bs->valid_key);
2248 }
2249
2250 int bdrv_set_key(BlockDriverState *bs, const char *key)
2251 {
2252 int ret;
2253 if (bs->backing_hd && bs->backing_hd->encrypted) {
2254 ret = bdrv_set_key(bs->backing_hd, key);
2255 if (ret < 0)
2256 return ret;
2257 if (!bs->encrypted)
2258 return 0;
2259 }
2260 if (!bs->encrypted) {
2261 return -EINVAL;
2262 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2263 return -ENOMEDIUM;
2264 }
2265 ret = bs->drv->bdrv_set_key(bs, key);
2266 if (ret < 0) {
2267 bs->valid_key = 0;
2268 } else if (!bs->valid_key) {
2269 bs->valid_key = 1;
2270 /* call the change callback now, we skipped it on open */
2271 bdrv_dev_change_media_cb(bs, true);
2272 }
2273 return ret;
2274 }
2275
2276 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2277 {
2278 if (!bs->drv) {
2279 buf[0] = '\0';
2280 } else {
2281 pstrcpy(buf, buf_size, bs->drv->format_name);
2282 }
2283 }
2284
2285 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2286 void *opaque)
2287 {
2288 BlockDriver *drv;
2289
2290 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2291 it(opaque, drv->format_name);
2292 }
2293 }
2294
2295 BlockDriverState *bdrv_find(const char *name)
2296 {
2297 BlockDriverState *bs;
2298
2299 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2300 if (!strcmp(name, bs->device_name)) {
2301 return bs;
2302 }
2303 }
2304 return NULL;
2305 }
2306
2307 BlockDriverState *bdrv_next(BlockDriverState *bs)
2308 {
2309 if (!bs) {
2310 return QTAILQ_FIRST(&bdrv_states);
2311 }
2312 return QTAILQ_NEXT(bs, list);
2313 }
2314
2315 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2316 {
2317 BlockDriverState *bs;
2318
2319 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2320 it(opaque, bs);
2321 }
2322 }
2323
2324 const char *bdrv_get_device_name(BlockDriverState *bs)
2325 {
2326 return bs->device_name;
2327 }
2328
2329 void bdrv_flush_all(void)
2330 {
2331 BlockDriverState *bs;
2332
2333 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2334 if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
2335 bdrv_flush(bs);
2336 }
2337 }
2338 }
2339
2340 int bdrv_has_zero_init(BlockDriverState *bs)
2341 {
2342 assert(bs->drv);
2343
2344 if (bs->drv->bdrv_has_zero_init) {
2345 return bs->drv->bdrv_has_zero_init(bs);
2346 }
2347
2348 return 1;
2349 }
2350
2351 typedef struct BdrvCoIsAllocatedData {
2352 BlockDriverState *bs;
2353 int64_t sector_num;
2354 int nb_sectors;
2355 int *pnum;
2356 int ret;
2357 bool done;
2358 } BdrvCoIsAllocatedData;
2359
2360 /*
2361 * Returns true iff the specified sector is present in the disk image. Drivers
2362 * not implementing the functionality are assumed to not support backing files,
2363 * hence all their sectors are reported as allocated.
2364 *
2365 * If 'sector_num' is beyond the end of the disk image the return value is 0
2366 * and 'pnum' is set to 0.
2367 *
2368 * 'pnum' is set to the number of sectors (including and immediately following
2369 * the specified sector) that are known to be in the same
2370 * allocated/unallocated state.
2371 *
2372 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2373 * beyond the end of the disk image it will be clamped.
2374 */
2375 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2376 int nb_sectors, int *pnum)
2377 {
2378 int64_t n;
2379
2380 if (sector_num >= bs->total_sectors) {
2381 *pnum = 0;
2382 return 0;
2383 }
2384
2385 n = bs->total_sectors - sector_num;
2386 if (n < nb_sectors) {
2387 nb_sectors = n;
2388 }
2389
2390 if (!bs->drv->bdrv_co_is_allocated) {
2391 *pnum = nb_sectors;
2392 return 1;
2393 }
2394
2395 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2396 }
2397
2398 /* Coroutine wrapper for bdrv_is_allocated() */
2399 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2400 {
2401 BdrvCoIsAllocatedData *data = opaque;
2402 BlockDriverState *bs = data->bs;
2403
2404 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2405 data->pnum);
2406 data->done = true;
2407 }
2408
2409 /*
2410 * Synchronous wrapper around bdrv_co_is_allocated().
2411 *
2412 * See bdrv_co_is_allocated() for details.
2413 */
2414 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2415 int *pnum)
2416 {
2417 Coroutine *co;
2418 BdrvCoIsAllocatedData data = {
2419 .bs = bs,
2420 .sector_num = sector_num,
2421 .nb_sectors = nb_sectors,
2422 .pnum = pnum,
2423 .done = false,
2424 };
2425
2426 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2427 qemu_coroutine_enter(co, &data);
2428 while (!data.done) {
2429 qemu_aio_wait();
2430 }
2431 return data.ret;
2432 }
2433
2434 BlockInfoList *qmp_query_block(Error **errp)
2435 {
2436 BlockInfoList *head = NULL, *cur_item = NULL;
2437 BlockDriverState *bs;
2438
2439 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2440 BlockInfoList *info = g_malloc0(sizeof(*info));
2441
2442 info->value = g_malloc0(sizeof(*info->value));
2443 info->value->device = g_strdup(bs->device_name);
2444 info->value->type = g_strdup("unknown");
2445 info->value->locked = bdrv_dev_is_medium_locked(bs);
2446 info->value->removable = bdrv_dev_has_removable_media(bs);
2447
2448 if (bdrv_dev_has_removable_media(bs)) {
2449 info->value->has_tray_open = true;
2450 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2451 }
2452
2453 if (bdrv_iostatus_is_enabled(bs)) {
2454 info->value->has_io_status = true;
2455 info->value->io_status = bs->iostatus;
2456 }
2457
2458 if (bs->drv) {
2459 info->value->has_inserted = true;
2460 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2461 info->value->inserted->file = g_strdup(bs->filename);
2462 info->value->inserted->ro = bs->read_only;
2463 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2464 info->value->inserted->encrypted = bs->encrypted;
2465 if (bs->backing_file[0]) {
2466 info->value->inserted->has_backing_file = true;
2467 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2468 }
2469
2470 if (bs->io_limits_enabled) {
2471 info->value->inserted->bps =
2472 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2473 info->value->inserted->bps_rd =
2474 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2475 info->value->inserted->bps_wr =
2476 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2477 info->value->inserted->iops =
2478 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2479 info->value->inserted->iops_rd =
2480 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2481 info->value->inserted->iops_wr =
2482 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2483 }
2484 }
2485
2486 /* XXX: waiting for the qapi to support GSList */
2487 if (!cur_item) {
2488 head = cur_item = info;
2489 } else {
2490 cur_item->next = info;
2491 cur_item = info;
2492 }
2493 }
2494
2495 return head;
2496 }
2497
2498 /* Consider exposing this as a full fledged QMP command */
2499 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2500 {
2501 BlockStats *s;
2502
2503 s = g_malloc0(sizeof(*s));
2504
2505 if (bs->device_name[0]) {
2506 s->has_device = true;
2507 s->device = g_strdup(bs->device_name);
2508 }
2509
2510 s->stats = g_malloc0(sizeof(*s->stats));
2511 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2512 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2513 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2514 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2515 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2516 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2517 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2518 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2519 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2520
2521 if (bs->file) {
2522 s->has_parent = true;
2523 s->parent = qmp_query_blockstat(bs->file, NULL);
2524 }
2525
2526 return s;
2527 }
2528
2529 BlockStatsList *qmp_query_blockstats(Error **errp)
2530 {
2531 BlockStatsList *head = NULL, *cur_item = NULL;
2532 BlockDriverState *bs;
2533
2534 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2535 BlockStatsList *info = g_malloc0(sizeof(*info));
2536 info->value = qmp_query_blockstat(bs, NULL);
2537
2538 /* XXX: waiting for the qapi to support GSList */
2539 if (!cur_item) {
2540 head = cur_item = info;
2541 } else {
2542 cur_item->next = info;
2543 cur_item = info;
2544 }
2545 }
2546
2547 return head;
2548 }
2549
2550 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2551 {
2552 if (bs->backing_hd && bs->backing_hd->encrypted)
2553 return bs->backing_file;
2554 else if (bs->encrypted)
2555 return bs->filename;
2556 else
2557 return NULL;
2558 }
2559
2560 void bdrv_get_backing_filename(BlockDriverState *bs,
2561 char *filename, int filename_size)
2562 {
2563 pstrcpy(filename, filename_size, bs->backing_file);
2564 }
2565
2566 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2567 const uint8_t *buf, int nb_sectors)
2568 {
2569 BlockDriver *drv = bs->drv;
2570 if (!drv)
2571 return -ENOMEDIUM;
2572 if (!drv->bdrv_write_compressed)
2573 return -ENOTSUP;
2574 if (bdrv_check_request(bs, sector_num, nb_sectors))
2575 return -EIO;
2576
2577 if (bs->dirty_bitmap) {
2578 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2579 }
2580
2581 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2582 }
2583
2584 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2585 {
2586 BlockDriver *drv = bs->drv;
2587 if (!drv)
2588 return -ENOMEDIUM;
2589 if (!drv->bdrv_get_info)
2590 return -ENOTSUP;
2591 memset(bdi, 0, sizeof(*bdi));
2592 return drv->bdrv_get_info(bs, bdi);
2593 }
2594
2595 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2596 int64_t pos, int size)
2597 {
2598 BlockDriver *drv = bs->drv;
2599 if (!drv)
2600 return -ENOMEDIUM;
2601 if (drv->bdrv_save_vmstate)
2602 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2603 if (bs->file)
2604 return bdrv_save_vmstate(bs->file, buf, pos, size);
2605 return -ENOTSUP;
2606 }
2607
2608 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2609 int64_t pos, int size)
2610 {
2611 BlockDriver *drv = bs->drv;
2612 if (!drv)
2613 return -ENOMEDIUM;
2614 if (drv->bdrv_load_vmstate)
2615 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2616 if (bs->file)
2617 return bdrv_load_vmstate(bs->file, buf, pos, size);
2618 return -ENOTSUP;
2619 }
2620
2621 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2622 {
2623 BlockDriver *drv = bs->drv;
2624
2625 if (!drv || !drv->bdrv_debug_event) {
2626 return;
2627 }
2628
2629 return drv->bdrv_debug_event(bs, event);
2630
2631 }
2632
2633 /**************************************************************/
2634 /* handling of snapshots */
2635
2636 int bdrv_can_snapshot(BlockDriverState *bs)
2637 {
2638 BlockDriver *drv = bs->drv;
2639 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2640 return 0;
2641 }
2642
2643 if (!drv->bdrv_snapshot_create) {
2644 if (bs->file != NULL) {
2645 return bdrv_can_snapshot(bs->file);
2646 }
2647 return 0;
2648 }
2649
2650 return 1;
2651 }
2652
2653 int bdrv_is_snapshot(BlockDriverState *bs)
2654 {
2655 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2656 }
2657
2658 BlockDriverState *bdrv_snapshots(void)
2659 {
2660 BlockDriverState *bs;
2661
2662 if (bs_snapshots) {
2663 return bs_snapshots;
2664 }
2665
2666 bs = NULL;
2667 while ((bs = bdrv_next(bs))) {
2668 if (bdrv_can_snapshot(bs)) {
2669 bs_snapshots = bs;
2670 return bs;
2671 }
2672 }
2673 return NULL;
2674 }
2675
2676 int bdrv_snapshot_create(BlockDriverState *bs,
2677 QEMUSnapshotInfo *sn_info)
2678 {
2679 BlockDriver *drv = bs->drv;
2680 if (!drv)
2681 return -ENOMEDIUM;
2682 if (drv->bdrv_snapshot_create)
2683 return drv->bdrv_snapshot_create(bs, sn_info);
2684 if (bs->file)
2685 return bdrv_snapshot_create(bs->file, sn_info);
2686 return -ENOTSUP;
2687 }
2688
2689 int bdrv_snapshot_goto(BlockDriverState *bs,
2690 const char *snapshot_id)
2691 {
2692 BlockDriver *drv = bs->drv;
2693 int ret, open_ret;
2694
2695 if (!drv)
2696 return -ENOMEDIUM;
2697 if (drv->bdrv_snapshot_goto)
2698 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2699
2700 if (bs->file) {
2701 drv->bdrv_close(bs);
2702 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2703 open_ret = drv->bdrv_open(bs, bs->open_flags);
2704 if (open_ret < 0) {
2705 bdrv_delete(bs->file);
2706 bs->drv = NULL;
2707 return open_ret;
2708 }
2709 return ret;
2710 }
2711
2712 return -ENOTSUP;
2713 }
2714
2715 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2716 {
2717 BlockDriver *drv = bs->drv;
2718 if (!drv)
2719 return -ENOMEDIUM;
2720 if (drv->bdrv_snapshot_delete)
2721 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2722 if (bs->file)
2723 return bdrv_snapshot_delete(bs->file, snapshot_id);
2724 return -ENOTSUP;
2725 }
2726
2727 int bdrv_snapshot_list(BlockDriverState *bs,
2728 QEMUSnapshotInfo **psn_info)
2729 {
2730 BlockDriver *drv = bs->drv;
2731 if (!drv)
2732 return -ENOMEDIUM;
2733 if (drv->bdrv_snapshot_list)
2734 return drv->bdrv_snapshot_list(bs, psn_info);
2735 if (bs->file)
2736 return bdrv_snapshot_list(bs->file, psn_info);
2737 return -ENOTSUP;
2738 }
2739
2740 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2741 const char *snapshot_name)
2742 {
2743 BlockDriver *drv = bs->drv;
2744 if (!drv) {
2745 return -ENOMEDIUM;
2746 }
2747 if (!bs->read_only) {
2748 return -EINVAL;
2749 }
2750 if (drv->bdrv_snapshot_load_tmp) {
2751 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2752 }
2753 return -ENOTSUP;
2754 }
2755
2756 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2757 const char *backing_file)
2758 {
2759 if (!bs->drv) {
2760 return NULL;
2761 }
2762
2763 if (bs->backing_hd) {
2764 if (strcmp(bs->backing_file, backing_file) == 0) {
2765 return bs->backing_hd;
2766 } else {
2767 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2768 }
2769 }
2770
2771 return NULL;
2772 }
2773
2774 #define NB_SUFFIXES 4
2775
2776 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2777 {
2778 static const char suffixes[NB_SUFFIXES] = "KMGT";
2779 int64_t base;
2780 int i;
2781
2782 if (size <= 999) {
2783 snprintf(buf, buf_size, "%" PRId64, size);
2784 } else {
2785 base = 1024;
2786 for(i = 0; i < NB_SUFFIXES; i++) {
2787 if (size < (10 * base)) {
2788 snprintf(buf, buf_size, "%0.1f%c",
2789 (double)size / base,
2790 suffixes[i]);
2791 break;
2792 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2793 snprintf(buf, buf_size, "%" PRId64 "%c",
2794 ((size + (base >> 1)) / base),
2795 suffixes[i]);
2796 break;
2797 }
2798 base = base * 1024;
2799 }
2800 }
2801 return buf;
2802 }
2803
2804 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2805 {
2806 char buf1[128], date_buf[128], clock_buf[128];
2807 #ifdef _WIN32
2808 struct tm *ptm;
2809 #else
2810 struct tm tm;
2811 #endif
2812 time_t ti;
2813 int64_t secs;
2814
2815 if (!sn) {
2816 snprintf(buf, buf_size,
2817 "%-10s%-20s%7s%20s%15s",
2818 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2819 } else {
2820 ti = sn->date_sec;
2821 #ifdef _WIN32
2822 ptm = localtime(&ti);
2823 strftime(date_buf, sizeof(date_buf),
2824 "%Y-%m-%d %H:%M:%S", ptm);
2825 #else
2826 localtime_r(&ti, &tm);
2827 strftime(date_buf, sizeof(date_buf),
2828 "%Y-%m-%d %H:%M:%S", &tm);
2829 #endif
2830 secs = sn->vm_clock_nsec / 1000000000;
2831 snprintf(clock_buf, sizeof(clock_buf),
2832 "%02d:%02d:%02d.%03d",
2833 (int)(secs / 3600),
2834 (int)((secs / 60) % 60),
2835 (int)(secs % 60),
2836 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2837 snprintf(buf, buf_size,
2838 "%-10s%-20s%7s%20s%15s",
2839 sn->id_str, sn->name,
2840 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2841 date_buf,
2842 clock_buf);
2843 }
2844 return buf;
2845 }
2846
2847 /**************************************************************/
2848 /* async I/Os */
2849
2850 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2851 QEMUIOVector *qiov, int nb_sectors,
2852 BlockDriverCompletionFunc *cb, void *opaque)
2853 {
2854 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2855
2856 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2857 cb, opaque, false);
2858 }
2859
2860 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2861 QEMUIOVector *qiov, int nb_sectors,
2862 BlockDriverCompletionFunc *cb, void *opaque)
2863 {
2864 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2865
2866 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2867 cb, opaque, true);
2868 }
2869
2870
2871 typedef struct MultiwriteCB {
2872 int error;
2873 int num_requests;
2874 int num_callbacks;
2875 struct {
2876 BlockDriverCompletionFunc *cb;
2877 void *opaque;
2878 QEMUIOVector *free_qiov;
2879 } callbacks[];
2880 } MultiwriteCB;
2881
2882 static void multiwrite_user_cb(MultiwriteCB *mcb)
2883 {
2884 int i;
2885
2886 for (i = 0; i < mcb->num_callbacks; i++) {
2887 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2888 if (mcb->callbacks[i].free_qiov) {
2889 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2890 }
2891 g_free(mcb->callbacks[i].free_qiov);
2892 }
2893 }
2894
2895 static void multiwrite_cb(void *opaque, int ret)
2896 {
2897 MultiwriteCB *mcb = opaque;
2898
2899 trace_multiwrite_cb(mcb, ret);
2900
2901 if (ret < 0 && !mcb->error) {
2902 mcb->error = ret;
2903 }
2904
2905 mcb->num_requests--;
2906 if (mcb->num_requests == 0) {
2907 multiwrite_user_cb(mcb);
2908 g_free(mcb);
2909 }
2910 }
2911
2912 static int multiwrite_req_compare(const void *a, const void *b)
2913 {
2914 const BlockRequest *req1 = a, *req2 = b;
2915
2916 /*
2917 * Note that we can't simply subtract req2->sector from req1->sector
2918 * here as that could overflow the return value.
2919 */
2920 if (req1->sector > req2->sector) {
2921 return 1;
2922 } else if (req1->sector < req2->sector) {
2923 return -1;
2924 } else {
2925 return 0;
2926 }
2927 }
2928
2929 /*
2930 * Takes a bunch of requests and tries to merge them. Returns the number of
2931 * requests that remain after merging.
2932 */
2933 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2934 int num_reqs, MultiwriteCB *mcb)
2935 {
2936 int i, outidx;
2937
2938 // Sort requests by start sector
2939 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2940
2941 // Check if adjacent requests touch the same clusters. If so, combine them,
2942 // filling up gaps with zero sectors.
2943 outidx = 0;
2944 for (i = 1; i < num_reqs; i++) {
2945 int merge = 0;
2946 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2947
2948 // Handle exactly sequential writes and overlapping writes.
2949 if (reqs[i].sector <= oldreq_last) {
2950 merge = 1;
2951 }
2952
2953 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2954 merge = 0;
2955 }
2956
2957 if (merge) {
2958 size_t size;
2959 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
2960 qemu_iovec_init(qiov,
2961 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2962
2963 // Add the first request to the merged one. If the requests are
2964 // overlapping, drop the last sectors of the first request.
2965 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2966 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2967
2968 // We should need to add any zeros between the two requests
2969 assert (reqs[i].sector <= oldreq_last);
2970
2971 // Add the second request
2972 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2973
2974 reqs[outidx].nb_sectors = qiov->size >> 9;
2975 reqs[outidx].qiov = qiov;
2976
2977 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2978 } else {
2979 outidx++;
2980 reqs[outidx].sector = reqs[i].sector;
2981 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2982 reqs[outidx].qiov = reqs[i].qiov;
2983 }
2984 }
2985
2986 return outidx + 1;
2987 }
2988
2989 /*
2990 * Submit multiple AIO write requests at once.
2991 *
2992 * On success, the function returns 0 and all requests in the reqs array have
2993 * been submitted. In error case this function returns -1, and any of the
2994 * requests may or may not be submitted yet. In particular, this means that the
2995 * callback will be called for some of the requests, for others it won't. The
2996 * caller must check the error field of the BlockRequest to wait for the right
2997 * callbacks (if error != 0, no callback will be called).
2998 *
2999 * The implementation may modify the contents of the reqs array, e.g. to merge
3000 * requests. However, the fields opaque and error are left unmodified as they
3001 * are used to signal failure for a single request to the caller.
3002 */
3003 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3004 {
3005 MultiwriteCB *mcb;
3006 int i;
3007
3008 /* don't submit writes if we don't have a medium */
3009 if (bs->drv == NULL) {
3010 for (i = 0; i < num_reqs; i++) {
3011 reqs[i].error = -ENOMEDIUM;
3012 }
3013 return -1;
3014 }
3015
3016 if (num_reqs == 0) {
3017 return 0;
3018 }
3019
3020 // Create MultiwriteCB structure
3021 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3022 mcb->num_requests = 0;
3023 mcb->num_callbacks = num_reqs;
3024
3025 for (i = 0; i < num_reqs; i++) {
3026 mcb->callbacks[i].cb = reqs[i].cb;
3027 mcb->callbacks[i].opaque = reqs[i].opaque;
3028 }
3029
3030 // Check for mergable requests
3031 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3032
3033 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3034
3035 /* Run the aio requests. */
3036 mcb->num_requests = num_reqs;
3037 for (i = 0; i < num_reqs; i++) {
3038 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3039 reqs[i].nb_sectors, multiwrite_cb, mcb);
3040 }
3041
3042 return 0;
3043 }
3044
3045 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3046 {
3047 acb->pool->cancel(acb);
3048 }
3049
3050 /* block I/O throttling */
3051 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3052 bool is_write, double elapsed_time, uint64_t *wait)
3053 {
3054 uint64_t bps_limit = 0;
3055 double bytes_limit, bytes_base, bytes_res;
3056 double slice_time, wait_time;
3057
3058 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3059 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3060 } else if (bs->io_limits.bps[is_write]) {
3061 bps_limit = bs->io_limits.bps[is_write];
3062 } else {
3063 if (wait) {
3064 *wait = 0;
3065 }
3066
3067 return false;
3068 }
3069
3070 slice_time = bs->slice_end - bs->slice_start;
3071 slice_time /= (NANOSECONDS_PER_SECOND);
3072 bytes_limit = bps_limit * slice_time;
3073 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3074 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3075 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3076 }
3077
3078 /* bytes_base: the bytes of data which have been read/written; and
3079 * it is obtained from the history statistic info.
3080 * bytes_res: the remaining bytes of data which need to be read/written.
3081 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3082 * the total time for completing reading/writting all data.
3083 */
3084 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3085
3086 if (bytes_base + bytes_res <= bytes_limit) {
3087 if (wait) {
3088 *wait = 0;
3089 }
3090
3091 return false;
3092 }
3093
3094 /* Calc approx time to dispatch */
3095 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3096
3097 /* When the I/O rate at runtime exceeds the limits,
3098 * bs->slice_end need to be extended in order that the current statistic
3099 * info can be kept until the timer fire, so it is increased and tuned
3100 * based on the result of experiment.
3101 */
3102 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3103 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3104 if (wait) {
3105 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3106 }
3107
3108 return true;
3109 }
3110
3111 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3112 double elapsed_time, uint64_t *wait)
3113 {
3114 uint64_t iops_limit = 0;
3115 double ios_limit, ios_base;
3116 double slice_time, wait_time;
3117
3118 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3119 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3120 } else if (bs->io_limits.iops[is_write]) {
3121 iops_limit = bs->io_limits.iops[is_write];
3122 } else {
3123 if (wait) {
3124 *wait = 0;
3125 }
3126
3127 return false;
3128 }
3129
3130 slice_time = bs->slice_end - bs->slice_start;
3131 slice_time /= (NANOSECONDS_PER_SECOND);
3132 ios_limit = iops_limit * slice_time;
3133 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3134 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3135 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3136 }
3137
3138 if (ios_base + 1 <= ios_limit) {
3139 if (wait) {
3140 *wait = 0;
3141 }
3142
3143 return false;
3144 }
3145
3146 /* Calc approx time to dispatch */
3147 wait_time = (ios_base + 1) / iops_limit;
3148 if (wait_time > elapsed_time) {
3149 wait_time = wait_time - elapsed_time;
3150 } else {
3151 wait_time = 0;
3152 }
3153
3154 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3155 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3156 if (wait) {
3157 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3158 }
3159
3160 return true;
3161 }
3162
3163 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3164 bool is_write, int64_t *wait)
3165 {
3166 int64_t now, max_wait;
3167 uint64_t bps_wait = 0, iops_wait = 0;
3168 double elapsed_time;
3169 int bps_ret, iops_ret;
3170
3171 now = qemu_get_clock_ns(vm_clock);
3172 if ((bs->slice_start < now)
3173 && (bs->slice_end > now)) {
3174 bs->slice_end = now + bs->slice_time;
3175 } else {
3176 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3177 bs->slice_start = now;
3178 bs->slice_end = now + bs->slice_time;
3179
3180 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3181 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3182
3183 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3184 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3185 }
3186
3187 elapsed_time = now - bs->slice_start;
3188 elapsed_time /= (NANOSECONDS_PER_SECOND);
3189
3190 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3191 is_write, elapsed_time, &bps_wait);
3192 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3193 elapsed_time, &iops_wait);
3194 if (bps_ret || iops_ret) {
3195 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3196 if (wait) {
3197 *wait = max_wait;
3198 }
3199
3200 now = qemu_get_clock_ns(vm_clock);
3201 if (bs->slice_end < now + max_wait) {
3202 bs->slice_end = now + max_wait;
3203 }
3204
3205 return true;
3206 }
3207
3208 if (wait) {
3209 *wait = 0;
3210 }
3211
3212 return false;
3213 }
3214
3215 /**************************************************************/
3216 /* async block device emulation */
3217
3218 typedef struct BlockDriverAIOCBSync {
3219 BlockDriverAIOCB common;
3220 QEMUBH *bh;
3221 int ret;
3222 /* vector translation state */
3223 QEMUIOVector *qiov;
3224 uint8_t *bounce;
3225 int is_write;
3226 } BlockDriverAIOCBSync;
3227
3228 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3229 {
3230 BlockDriverAIOCBSync *acb =
3231 container_of(blockacb, BlockDriverAIOCBSync, common);
3232 qemu_bh_delete(acb->bh);
3233 acb->bh = NULL;
3234 qemu_aio_release(acb);
3235 }
3236
3237 static AIOPool bdrv_em_aio_pool = {
3238 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3239 .cancel = bdrv_aio_cancel_em,
3240 };
3241
3242 static void bdrv_aio_bh_cb(void *opaque)
3243 {
3244 BlockDriverAIOCBSync *acb = opaque;
3245
3246 if (!acb->is_write)
3247 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3248 qemu_vfree(acb->bounce);
3249 acb->common.cb(acb->common.opaque, acb->ret);
3250 qemu_bh_delete(acb->bh);
3251 acb->bh = NULL;
3252 qemu_aio_release(acb);
3253 }
3254
3255 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3256 int64_t sector_num,
3257 QEMUIOVector *qiov,
3258 int nb_sectors,
3259 BlockDriverCompletionFunc *cb,
3260 void *opaque,
3261 int is_write)
3262
3263 {
3264 BlockDriverAIOCBSync *acb;
3265
3266 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3267 acb->is_write = is_write;
3268 acb->qiov = qiov;
3269 acb->bounce = qemu_blockalign(bs, qiov->size);
3270 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3271
3272 if (is_write) {
3273 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3274 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3275 } else {
3276 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3277 }
3278
3279 qemu_bh_schedule(acb->bh);
3280
3281 return &acb->common;
3282 }
3283
3284 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3285 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3286 BlockDriverCompletionFunc *cb, void *opaque)
3287 {
3288 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3289 }
3290
3291 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3292 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3293 BlockDriverCompletionFunc *cb, void *opaque)
3294 {
3295 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3296 }
3297
3298
3299 typedef struct BlockDriverAIOCBCoroutine {
3300 BlockDriverAIOCB common;
3301 BlockRequest req;
3302 bool is_write;
3303 QEMUBH* bh;
3304 } BlockDriverAIOCBCoroutine;
3305
3306 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3307 {
3308 qemu_aio_flush();
3309 }
3310
3311 static AIOPool bdrv_em_co_aio_pool = {
3312 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3313 .cancel = bdrv_aio_co_cancel_em,
3314 };
3315
3316 static void bdrv_co_em_bh(void *opaque)
3317 {
3318 BlockDriverAIOCBCoroutine *acb = opaque;
3319
3320 acb->common.cb(acb->common.opaque, acb->req.error);
3321 qemu_bh_delete(acb->bh);
3322 qemu_aio_release(acb);
3323 }
3324
3325 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3326 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3327 {
3328 BlockDriverAIOCBCoroutine *acb = opaque;
3329 BlockDriverState *bs = acb->common.bs;
3330
3331 if (!acb->is_write) {
3332 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3333 acb->req.nb_sectors, acb->req.qiov, 0);
3334 } else {
3335 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3336 acb->req.nb_sectors, acb->req.qiov, 0);
3337 }
3338
3339 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3340 qemu_bh_schedule(acb->bh);
3341 }
3342
3343 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3344 int64_t sector_num,
3345 QEMUIOVector *qiov,
3346 int nb_sectors,
3347 BlockDriverCompletionFunc *cb,
3348 void *opaque,
3349 bool is_write)
3350 {
3351 Coroutine *co;
3352 BlockDriverAIOCBCoroutine *acb;
3353
3354 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3355 acb->req.sector = sector_num;
3356 acb->req.nb_sectors = nb_sectors;
3357 acb->req.qiov = qiov;
3358 acb->is_write = is_write;
3359
3360 co = qemu_coroutine_create(bdrv_co_do_rw);
3361 qemu_coroutine_enter(co, acb);
3362
3363 return &acb->common;
3364 }
3365
3366 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3367 {
3368 BlockDriverAIOCBCoroutine *acb = opaque;
3369 BlockDriverState *bs = acb->common.bs;
3370
3371 acb->req.error = bdrv_co_flush(bs);
3372 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3373 qemu_bh_schedule(acb->bh);
3374 }
3375
3376 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3377 BlockDriverCompletionFunc *cb, void *opaque)
3378 {
3379 trace_bdrv_aio_flush(bs, opaque);
3380
3381 Coroutine *co;
3382 BlockDriverAIOCBCoroutine *acb;
3383
3384 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3385 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3386 qemu_coroutine_enter(co, acb);
3387
3388 return &acb->common;
3389 }
3390
3391 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3392 {
3393 BlockDriverAIOCBCoroutine *acb = opaque;
3394 BlockDriverState *bs = acb->common.bs;
3395
3396 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3397 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3398 qemu_bh_schedule(acb->bh);
3399 }
3400
3401 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3402 int64_t sector_num, int nb_sectors,
3403 BlockDriverCompletionFunc *cb, void *opaque)
3404 {
3405 Coroutine *co;
3406 BlockDriverAIOCBCoroutine *acb;
3407
3408 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3409
3410 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3411 acb->req.sector = sector_num;
3412 acb->req.nb_sectors = nb_sectors;
3413 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3414 qemu_coroutine_enter(co, acb);
3415
3416 return &acb->common;
3417 }
3418
3419 void bdrv_init(void)
3420 {
3421 module_call_init(MODULE_INIT_BLOCK);
3422 }
3423
3424 void bdrv_init_with_whitelist(void)
3425 {
3426 use_bdrv_whitelist = 1;
3427 bdrv_init();
3428 }
3429
3430 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3431 BlockDriverCompletionFunc *cb, void *opaque)
3432 {
3433 BlockDriverAIOCB *acb;
3434
3435 if (pool->free_aiocb) {
3436 acb = pool->free_aiocb;
3437 pool->free_aiocb = acb->next;
3438 } else {
3439 acb = g_malloc0(pool->aiocb_size);
3440 acb->pool = pool;
3441 }
3442 acb->bs = bs;
3443 acb->cb = cb;
3444 acb->opaque = opaque;
3445 return acb;
3446 }
3447
3448 void qemu_aio_release(void *p)
3449 {
3450 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3451 AIOPool *pool = acb->pool;
3452 acb->next = pool->free_aiocb;
3453 pool->free_aiocb = acb;
3454 }
3455
3456 /**************************************************************/
3457 /* Coroutine block device emulation */
3458
3459 typedef struct CoroutineIOCompletion {
3460 Coroutine *coroutine;
3461 int ret;
3462 } CoroutineIOCompletion;
3463
3464 static void bdrv_co_io_em_complete(void *opaque, int ret)
3465 {
3466 CoroutineIOCompletion *co = opaque;
3467
3468 co->ret = ret;
3469 qemu_coroutine_enter(co->coroutine, NULL);
3470 }
3471
3472 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3473 int nb_sectors, QEMUIOVector *iov,
3474 bool is_write)
3475 {
3476 CoroutineIOCompletion co = {
3477 .coroutine = qemu_coroutine_self(),
3478 };
3479 BlockDriverAIOCB *acb;
3480
3481 if (is_write) {
3482 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3483 bdrv_co_io_em_complete, &co);
3484 } else {
3485 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3486 bdrv_co_io_em_complete, &co);
3487 }
3488
3489 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3490 if (!acb) {
3491 return -EIO;
3492 }
3493 qemu_coroutine_yield();
3494
3495 return co.ret;
3496 }
3497
3498 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3499 int64_t sector_num, int nb_sectors,
3500 QEMUIOVector *iov)
3501 {
3502 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3503 }
3504
3505 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3506 int64_t sector_num, int nb_sectors,
3507 QEMUIOVector *iov)
3508 {
3509 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3510 }
3511
3512 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3513 {
3514 RwCo *rwco = opaque;
3515
3516 rwco->ret = bdrv_co_flush(rwco->bs);
3517 }
3518
3519 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3520 {
3521 int ret;
3522
3523 if (!bs->drv) {
3524 return 0;
3525 }
3526
3527 /* Write back cached data to the OS even with cache=unsafe */
3528 if (bs->drv->bdrv_co_flush_to_os) {
3529 ret = bs->drv->bdrv_co_flush_to_os(bs);
3530 if (ret < 0) {
3531 return ret;
3532 }
3533 }
3534
3535 /* But don't actually force it to the disk with cache=unsafe */
3536 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3537 return 0;
3538 }
3539
3540 if (bs->drv->bdrv_co_flush_to_disk) {
3541 return bs->drv->bdrv_co_flush_to_disk(bs);
3542 } else if (bs->drv->bdrv_aio_flush) {
3543 BlockDriverAIOCB *acb;
3544 CoroutineIOCompletion co = {
3545 .coroutine = qemu_coroutine_self(),
3546 };
3547
3548 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3549 if (acb == NULL) {
3550 return -EIO;
3551 } else {
3552 qemu_coroutine_yield();
3553 return co.ret;
3554 }
3555 } else {
3556 /*
3557 * Some block drivers always operate in either writethrough or unsafe
3558 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3559 * know how the server works (because the behaviour is hardcoded or
3560 * depends on server-side configuration), so we can't ensure that
3561 * everything is safe on disk. Returning an error doesn't work because
3562 * that would break guests even if the server operates in writethrough
3563 * mode.
3564 *
3565 * Let's hope the user knows what he's doing.
3566 */
3567 return 0;
3568 }
3569 }
3570
3571 void bdrv_invalidate_cache(BlockDriverState *bs)
3572 {
3573 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3574 bs->drv->bdrv_invalidate_cache(bs);
3575 }
3576 }
3577
3578 void bdrv_invalidate_cache_all(void)
3579 {
3580 BlockDriverState *bs;
3581
3582 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3583 bdrv_invalidate_cache(bs);
3584 }
3585 }
3586
3587 int bdrv_flush(BlockDriverState *bs)
3588 {
3589 Coroutine *co;
3590 RwCo rwco = {
3591 .bs = bs,
3592 .ret = NOT_DONE,
3593 };
3594
3595 if (qemu_in_coroutine()) {
3596 /* Fast-path if already in coroutine context */
3597 bdrv_flush_co_entry(&rwco);
3598 } else {
3599 co = qemu_coroutine_create(bdrv_flush_co_entry);
3600 qemu_coroutine_enter(co, &rwco);
3601 while (rwco.ret == NOT_DONE) {
3602 qemu_aio_wait();
3603 }
3604 }
3605
3606 return rwco.ret;
3607 }
3608
3609 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3610 {
3611 RwCo *rwco = opaque;
3612
3613 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3614 }
3615
3616 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3617 int nb_sectors)
3618 {
3619 if (!bs->drv) {
3620 return -ENOMEDIUM;
3621 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3622 return -EIO;
3623 } else if (bs->read_only) {
3624 return -EROFS;
3625 } else if (bs->drv->bdrv_co_discard) {
3626 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3627 } else if (bs->drv->bdrv_aio_discard) {
3628 BlockDriverAIOCB *acb;
3629 CoroutineIOCompletion co = {
3630 .coroutine = qemu_coroutine_self(),
3631 };
3632
3633 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3634 bdrv_co_io_em_complete, &co);
3635 if (acb == NULL) {
3636 return -EIO;
3637 } else {
3638 qemu_coroutine_yield();
3639 return co.ret;
3640 }
3641 } else {
3642 return 0;
3643 }
3644 }
3645
3646 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3647 {
3648 Coroutine *co;
3649 RwCo rwco = {
3650 .bs = bs,
3651 .sector_num = sector_num,
3652 .nb_sectors = nb_sectors,
3653 .ret = NOT_DONE,
3654 };
3655
3656 if (qemu_in_coroutine()) {
3657 /* Fast-path if already in coroutine context */
3658 bdrv_discard_co_entry(&rwco);
3659 } else {
3660 co = qemu_coroutine_create(bdrv_discard_co_entry);
3661 qemu_coroutine_enter(co, &rwco);
3662 while (rwco.ret == NOT_DONE) {
3663 qemu_aio_wait();
3664 }
3665 }
3666
3667 return rwco.ret;
3668 }
3669
3670 /**************************************************************/
3671 /* removable device support */
3672
3673 /**
3674 * Return TRUE if the media is present
3675 */
3676 int bdrv_is_inserted(BlockDriverState *bs)
3677 {
3678 BlockDriver *drv = bs->drv;
3679
3680 if (!drv)
3681 return 0;
3682 if (!drv->bdrv_is_inserted)
3683 return 1;
3684 return drv->bdrv_is_inserted(bs);
3685 }
3686
3687 /**
3688 * Return whether the media changed since the last call to this
3689 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3690 */
3691 int bdrv_media_changed(BlockDriverState *bs)
3692 {
3693 BlockDriver *drv = bs->drv;
3694
3695 if (drv && drv->bdrv_media_changed) {
3696 return drv->bdrv_media_changed(bs);
3697 }
3698 return -ENOTSUP;
3699 }
3700
3701 /**
3702 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3703 */
3704 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3705 {
3706 BlockDriver *drv = bs->drv;
3707
3708 if (drv && drv->bdrv_eject) {
3709 drv->bdrv_eject(bs, eject_flag);
3710 }
3711
3712 if (bs->device_name[0] != '\0') {
3713 bdrv_emit_qmp_eject_event(bs, eject_flag);
3714 }
3715 }
3716
3717 /**
3718 * Lock or unlock the media (if it is locked, the user won't be able
3719 * to eject it manually).
3720 */
3721 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3722 {
3723 BlockDriver *drv = bs->drv;
3724
3725 trace_bdrv_lock_medium(bs, locked);
3726
3727 if (drv && drv->bdrv_lock_medium) {
3728 drv->bdrv_lock_medium(bs, locked);
3729 }
3730 }
3731
3732 /* needed for generic scsi interface */
3733
3734 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3735 {
3736 BlockDriver *drv = bs->drv;
3737
3738 if (drv && drv->bdrv_ioctl)
3739 return drv->bdrv_ioctl(bs, req, buf);
3740 return -ENOTSUP;
3741 }
3742
3743 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3744 unsigned long int req, void *buf,
3745 BlockDriverCompletionFunc *cb, void *opaque)
3746 {
3747 BlockDriver *drv = bs->drv;
3748
3749 if (drv && drv->bdrv_aio_ioctl)
3750 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3751 return NULL;
3752 }
3753
3754 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3755 {
3756 bs->buffer_alignment = align;
3757 }
3758
3759 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3760 {
3761 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3762 }
3763
3764 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3765 {
3766 int64_t bitmap_size;
3767
3768 bs->dirty_count = 0;
3769 if (enable) {
3770 if (!bs->dirty_bitmap) {
3771 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3772 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3773 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3774
3775 bs->dirty_bitmap = g_malloc0(bitmap_size);
3776 }
3777 } else {
3778 if (bs->dirty_bitmap) {
3779 g_free(bs->dirty_bitmap);
3780 bs->dirty_bitmap = NULL;
3781 }
3782 }
3783 }
3784
3785 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3786 {
3787 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3788
3789 if (bs->dirty_bitmap &&
3790 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3791 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3792 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3793 } else {
3794 return 0;
3795 }
3796 }
3797
3798 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3799 int nr_sectors)
3800 {
3801 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3802 }
3803
3804 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3805 {
3806 return bs->dirty_count;
3807 }
3808
3809 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3810 {
3811 assert(bs->in_use != in_use);
3812 bs->in_use = in_use;
3813 }
3814
3815 int bdrv_in_use(BlockDriverState *bs)
3816 {
3817 return bs->in_use;
3818 }
3819
3820 void bdrv_iostatus_enable(BlockDriverState *bs)
3821 {
3822 bs->iostatus_enabled = true;
3823 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3824 }
3825
3826 /* The I/O status is only enabled if the drive explicitly
3827 * enables it _and_ the VM is configured to stop on errors */
3828 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3829 {
3830 return (bs->iostatus_enabled &&
3831 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3832 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3833 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3834 }
3835
3836 void bdrv_iostatus_disable(BlockDriverState *bs)
3837 {
3838 bs->iostatus_enabled = false;
3839 }
3840
3841 void bdrv_iostatus_reset(BlockDriverState *bs)
3842 {
3843 if (bdrv_iostatus_is_enabled(bs)) {
3844 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3845 }
3846 }
3847
3848 /* XXX: Today this is set by device models because it makes the implementation
3849 quite simple. However, the block layer knows about the error, so it's
3850 possible to implement this without device models being involved */
3851 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3852 {
3853 if (bdrv_iostatus_is_enabled(bs) &&
3854 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3855 assert(error >= 0);
3856 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
3857 BLOCK_DEVICE_IO_STATUS_FAILED;
3858 }
3859 }
3860
3861 void
3862 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3863 enum BlockAcctType type)
3864 {
3865 assert(type < BDRV_MAX_IOTYPE);
3866
3867 cookie->bytes = bytes;
3868 cookie->start_time_ns = get_clock();
3869 cookie->type = type;
3870 }
3871
3872 void
3873 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3874 {
3875 assert(cookie->type < BDRV_MAX_IOTYPE);
3876
3877 bs->nr_bytes[cookie->type] += cookie->bytes;
3878 bs->nr_ops[cookie->type]++;
3879 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
3880 }
3881
3882 int bdrv_img_create(const char *filename, const char *fmt,
3883 const char *base_filename, const char *base_fmt,
3884 char *options, uint64_t img_size, int flags)
3885 {
3886 QEMUOptionParameter *param = NULL, *create_options = NULL;
3887 QEMUOptionParameter *backing_fmt, *backing_file, *size;
3888 BlockDriverState *bs = NULL;
3889 BlockDriver *drv, *proto_drv;
3890 BlockDriver *backing_drv = NULL;
3891 int ret = 0;
3892
3893 /* Find driver and parse its options */
3894 drv = bdrv_find_format(fmt);
3895 if (!drv) {
3896 error_report("Unknown file format '%s'", fmt);
3897 ret = -EINVAL;
3898 goto out;
3899 }
3900
3901 proto_drv = bdrv_find_protocol(filename);
3902 if (!proto_drv) {
3903 error_report("Unknown protocol '%s'", filename);
3904 ret = -EINVAL;
3905 goto out;
3906 }
3907
3908 create_options = append_option_parameters(create_options,
3909 drv->create_options);
3910 create_options = append_option_parameters(create_options,
3911 proto_drv->create_options);
3912
3913 /* Create parameter list with default values */
3914 param = parse_option_parameters("", create_options, param);
3915
3916 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3917
3918 /* Parse -o options */
3919 if (options) {
3920 param = parse_option_parameters(options, create_options, param);
3921 if (param == NULL) {
3922 error_report("Invalid options for file format '%s'.", fmt);
3923 ret = -EINVAL;
3924 goto out;
3925 }
3926 }
3927
3928 if (base_filename) {
3929 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3930 base_filename)) {
3931 error_report("Backing file not supported for file format '%s'",
3932 fmt);
3933 ret = -EINVAL;
3934 goto out;
3935 }
3936 }
3937
3938 if (base_fmt) {
3939 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
3940 error_report("Backing file format not supported for file "
3941 "format '%s'", fmt);
3942 ret = -EINVAL;
3943 goto out;
3944 }
3945 }
3946
3947 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
3948 if (backing_file && backing_file->value.s) {
3949 if (!strcmp(filename, backing_file->value.s)) {
3950 error_report("Error: Trying to create an image with the "
3951 "same filename as the backing file");
3952 ret = -EINVAL;
3953 goto out;
3954 }
3955 }
3956
3957 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
3958 if (backing_fmt && backing_fmt->value.s) {
3959 backing_drv = bdrv_find_format(backing_fmt->value.s);
3960 if (!backing_drv) {
3961 error_report("Unknown backing file format '%s'",
3962 backing_fmt->value.s);
3963 ret = -EINVAL;
3964 goto out;
3965 }
3966 }
3967
3968 // The size for the image must always be specified, with one exception:
3969 // If we are using a backing file, we can obtain the size from there
3970 size = get_option_parameter(param, BLOCK_OPT_SIZE);
3971 if (size && size->value.n == -1) {
3972 if (backing_file && backing_file->value.s) {
3973 uint64_t size;
3974 char buf[32];
3975
3976 bs = bdrv_new("");
3977
3978 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
3979 if (ret < 0) {
3980 error_report("Could not open '%s'", backing_file->value.s);
3981 goto out;
3982 }
3983 bdrv_get_geometry(bs, &size);
3984 size *= 512;
3985
3986 snprintf(buf, sizeof(buf), "%" PRId64, size);
3987 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3988 } else {
3989 error_report("Image creation needs a size parameter");
3990 ret = -EINVAL;
3991 goto out;
3992 }
3993 }
3994
3995 printf("Formatting '%s', fmt=%s ", filename, fmt);
3996 print_option_parameters(param);
3997 puts("");
3998
3999 ret = bdrv_create(drv, filename, param);
4000
4001 if (ret < 0) {
4002 if (ret == -ENOTSUP) {
4003 error_report("Formatting or formatting option not supported for "
4004 "file format '%s'", fmt);
4005 } else if (ret == -EFBIG) {
4006 error_report("The image size is too large for file format '%s'",
4007 fmt);
4008 } else {
4009 error_report("%s: error while creating %s: %s", filename, fmt,
4010 strerror(-ret));
4011 }
4012 }
4013
4014 out:
4015 free_option_parameters(create_options);
4016 free_option_parameters(param);
4017
4018 if (bs) {
4019 bdrv_delete(bs);
4020 }
4021
4022 return ret;
4023 }
4024
4025 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4026 BlockDriverCompletionFunc *cb, void *opaque)
4027 {
4028 BlockJob *job;
4029
4030 if (bs->job || bdrv_in_use(bs)) {
4031 return NULL;
4032 }
4033 bdrv_set_in_use(bs, 1);
4034
4035 job = g_malloc0(job_type->instance_size);
4036 job->job_type = job_type;
4037 job->bs = bs;
4038 job->cb = cb;
4039 job->opaque = opaque;
4040 bs->job = job;
4041 return job;
4042 }
4043
4044 void block_job_complete(BlockJob *job, int ret)
4045 {
4046 BlockDriverState *bs = job->bs;
4047
4048 assert(bs->job == job);
4049 job->cb(job->opaque, ret);
4050 bs->job = NULL;
4051 g_free(job);
4052 bdrv_set_in_use(bs, 0);
4053 }
4054
4055 int block_job_set_speed(BlockJob *job, int64_t value)
4056 {
4057 if (!job->job_type->set_speed) {
4058 return -ENOTSUP;
4059 }
4060 return job->job_type->set_speed(job, value);
4061 }
4062
4063 void block_job_cancel(BlockJob *job)
4064 {
4065 job->cancelled = true;
4066 }
4067
4068 bool block_job_is_cancelled(BlockJob *job)
4069 {
4070 return job->cancelled;
4071 }
Patch queue for openSUSE QEMU