?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
main-loop: For tools, initialize timers as part of qemu_init_main_loop()
[qemu/kevin.git] / block.c
blob3621d11de0ab8b012e9e0b2dd229cc0ded5c38aa
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 } BdrvRequestFlags;
54
55 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
56 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
57 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
58 BlockDriverCompletionFunc *cb, void *opaque);
59 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
60 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
61 BlockDriverCompletionFunc *cb, void *opaque);
62 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
63 int64_t sector_num, int nb_sectors,
64 QEMUIOVector *iov);
65 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
66 int64_t sector_num, int nb_sectors,
67 QEMUIOVector *iov);
68 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
69 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
70 BdrvRequestFlags flags);
71 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
72 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
73 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
74 int64_t sector_num,
75 QEMUIOVector *qiov,
76 int nb_sectors,
77 BlockDriverCompletionFunc *cb,
78 void *opaque,
79 bool is_write);
80 static void coroutine_fn bdrv_co_do_rw(void *opaque);
81
82 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
83 bool is_write, double elapsed_time, uint64_t *wait);
84 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
85 double elapsed_time, uint64_t *wait);
86 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
87 bool is_write, int64_t *wait);
88
89 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
90 QTAILQ_HEAD_INITIALIZER(bdrv_states);
91
92 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
93 QLIST_HEAD_INITIALIZER(bdrv_drivers);
94
95 /* The device to use for VM snapshots */
96 static BlockDriverState *bs_snapshots;
97
98 /* If non-zero, use only whitelisted block drivers */
99 static int use_bdrv_whitelist;
100
101 #ifdef _WIN32
102 static int is_windows_drive_prefix(const char *filename)
103 {
104 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
105 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
106 filename[1] == ':');
107 }
108
109 int is_windows_drive(const char *filename)
110 {
111 if (is_windows_drive_prefix(filename) &&
112 filename[2] == '\0')
113 return 1;
114 if (strstart(filename, "\\\\.\\", NULL) ||
115 strstart(filename, "//./", NULL))
116 return 1;
117 return 0;
118 }
119 #endif
120
121 /* throttling disk I/O limits */
122 void bdrv_io_limits_disable(BlockDriverState *bs)
123 {
124 bs->io_limits_enabled = false;
125
126 while (qemu_co_queue_next(&bs->throttled_reqs));
127
128 if (bs->block_timer) {
129 qemu_del_timer(bs->block_timer);
130 qemu_free_timer(bs->block_timer);
131 bs->block_timer = NULL;
132 }
133
134 bs->slice_start = 0;
135 bs->slice_end = 0;
136 bs->slice_time = 0;
137 memset(&bs->io_base, 0, sizeof(bs->io_base));
138 }
139
140 static void bdrv_block_timer(void *opaque)
141 {
142 BlockDriverState *bs = opaque;
143
144 qemu_co_queue_next(&bs->throttled_reqs);
145 }
146
147 void bdrv_io_limits_enable(BlockDriverState *bs)
148 {
149 qemu_co_queue_init(&bs->throttled_reqs);
150 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
151 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
152 bs->slice_start = qemu_get_clock_ns(vm_clock);
153 bs->slice_end = bs->slice_start + bs->slice_time;
154 memset(&bs->io_base, 0, sizeof(bs->io_base));
155 bs->io_limits_enabled = true;
156 }
157
158 bool bdrv_io_limits_enabled(BlockDriverState *bs)
159 {
160 BlockIOLimit *io_limits = &bs->io_limits;
161 return io_limits->bps[BLOCK_IO_LIMIT_READ]
162 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
163 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
164 || io_limits->iops[BLOCK_IO_LIMIT_READ]
165 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
166 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
167 }
168
169 static void bdrv_io_limits_intercept(BlockDriverState *bs,
170 bool is_write, int nb_sectors)
171 {
172 int64_t wait_time = -1;
173
174 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
175 qemu_co_queue_wait(&bs->throttled_reqs);
176 }
177
178 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
179 * throttled requests will not be dequeued until the current request is
180 * allowed to be serviced. So if the current request still exceeds the
181 * limits, it will be inserted to the head. All requests followed it will
182 * be still in throttled_reqs queue.
183 */
184
185 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
186 qemu_mod_timer(bs->block_timer,
187 wait_time + qemu_get_clock_ns(vm_clock));
188 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
189 }
190
191 qemu_co_queue_next(&bs->throttled_reqs);
192 }
193
194 /* check if the path starts with "<protocol>:" */
195 static int path_has_protocol(const char *path)
196 {
197 #ifdef _WIN32
198 if (is_windows_drive(path) ||
199 is_windows_drive_prefix(path)) {
200 return 0;
201 }
202 #endif
203
204 return strchr(path, ':') != NULL;
205 }
206
207 int path_is_absolute(const char *path)
208 {
209 const char *p;
210 #ifdef _WIN32
211 /* specific case for names like: "\\.\d:" */
212 if (*path == '/' || *path == '\\')
213 return 1;
214 #endif
215 p = strchr(path, ':');
216 if (p)
217 p++;
218 else
219 p = path;
220 #ifdef _WIN32
221 return (*p == '/' || *p == '\\');
222 #else
223 return (*p == '/');
224 #endif
225 }
226
227 /* if filename is absolute, just copy it to dest. Otherwise, build a
228 path to it by considering it is relative to base_path. URL are
229 supported. */
230 void path_combine(char *dest, int dest_size,
231 const char *base_path,
232 const char *filename)
233 {
234 const char *p, *p1;
235 int len;
236
237 if (dest_size <= 0)
238 return;
239 if (path_is_absolute(filename)) {
240 pstrcpy(dest, dest_size, filename);
241 } else {
242 p = strchr(base_path, ':');
243 if (p)
244 p++;
245 else
246 p = base_path;
247 p1 = strrchr(base_path, '/');
248 #ifdef _WIN32
249 {
250 const char *p2;
251 p2 = strrchr(base_path, '\\');
252 if (!p1 || p2 > p1)
253 p1 = p2;
254 }
255 #endif
256 if (p1)
257 p1++;
258 else
259 p1 = base_path;
260 if (p1 > p)
261 p = p1;
262 len = p - base_path;
263 if (len > dest_size - 1)
264 len = dest_size - 1;
265 memcpy(dest, base_path, len);
266 dest[len] = '\0';
267 pstrcat(dest, dest_size, filename);
268 }
269 }
270
271 void bdrv_register(BlockDriver *bdrv)
272 {
273 /* Block drivers without coroutine functions need emulation */
274 if (!bdrv->bdrv_co_readv) {
275 bdrv->bdrv_co_readv = bdrv_co_readv_em;
276 bdrv->bdrv_co_writev = bdrv_co_writev_em;
277
278 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
279 * the block driver lacks aio we need to emulate that too.
280 */
281 if (!bdrv->bdrv_aio_readv) {
282 /* add AIO emulation layer */
283 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
284 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
285 }
286 }
287
288 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
289 }
290
291 /* create a new block device (by default it is empty) */
292 BlockDriverState *bdrv_new(const char *device_name)
293 {
294 BlockDriverState *bs;
295
296 bs = g_malloc0(sizeof(BlockDriverState));
297 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
298 if (device_name[0] != '\0') {
299 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
300 }
301 bdrv_iostatus_disable(bs);
302 return bs;
303 }
304
305 BlockDriver *bdrv_find_format(const char *format_name)
306 {
307 BlockDriver *drv1;
308 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
309 if (!strcmp(drv1->format_name, format_name)) {
310 return drv1;
311 }
312 }
313 return NULL;
314 }
315
316 static int bdrv_is_whitelisted(BlockDriver *drv)
317 {
318 static const char *whitelist[] = {
319 CONFIG_BDRV_WHITELIST
320 };
321 const char **p;
322
323 if (!whitelist[0])
324 return 1; /* no whitelist, anything goes */
325
326 for (p = whitelist; *p; p++) {
327 if (!strcmp(drv->format_name, *p)) {
328 return 1;
329 }
330 }
331 return 0;
332 }
333
334 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
335 {
336 BlockDriver *drv = bdrv_find_format(format_name);
337 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
338 }
339
340 int bdrv_create(BlockDriver *drv, const char* filename,
341 QEMUOptionParameter *options)
342 {
343 if (!drv->bdrv_create)
344 return -ENOTSUP;
345
346 return drv->bdrv_create(filename, options);
347 }
348
349 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
350 {
351 BlockDriver *drv;
352
353 drv = bdrv_find_protocol(filename);
354 if (drv == NULL) {
355 return -ENOENT;
356 }
357
358 return bdrv_create(drv, filename, options);
359 }
360
361 #ifdef _WIN32
362 void get_tmp_filename(char *filename, int size)
363 {
364 char temp_dir[MAX_PATH];
365
366 GetTempPath(MAX_PATH, temp_dir);
367 GetTempFileName(temp_dir, "qem", 0, filename);
368 }
369 #else
370 void get_tmp_filename(char *filename, int size)
371 {
372 int fd;
373 const char *tmpdir;
374 /* XXX: race condition possible */
375 tmpdir = getenv("TMPDIR");
376 if (!tmpdir)
377 tmpdir = "/tmp";
378 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
379 fd = mkstemp(filename);
380 close(fd);
381 }
382 #endif
383
384 /*
385 * Detect host devices. By convention, /dev/cdrom[N] is always
386 * recognized as a host CDROM.
387 */
388 static BlockDriver *find_hdev_driver(const char *filename)
389 {
390 int score_max = 0, score;
391 BlockDriver *drv = NULL, *d;
392
393 QLIST_FOREACH(d, &bdrv_drivers, list) {
394 if (d->bdrv_probe_device) {
395 score = d->bdrv_probe_device(filename);
396 if (score > score_max) {
397 score_max = score;
398 drv = d;
399 }
400 }
401 }
402
403 return drv;
404 }
405
406 BlockDriver *bdrv_find_protocol(const char *filename)
407 {
408 BlockDriver *drv1;
409 char protocol[128];
410 int len;
411 const char *p;
412
413 /* TODO Drivers without bdrv_file_open must be specified explicitly */
414
415 /*
416 * XXX(hch): we really should not let host device detection
417 * override an explicit protocol specification, but moving this
418 * later breaks access to device names with colons in them.
419 * Thanks to the brain-dead persistent naming schemes on udev-
420 * based Linux systems those actually are quite common.
421 */
422 drv1 = find_hdev_driver(filename);
423 if (drv1) {
424 return drv1;
425 }
426
427 if (!path_has_protocol(filename)) {
428 return bdrv_find_format("file");
429 }
430 p = strchr(filename, ':');
431 assert(p != NULL);
432 len = p - filename;
433 if (len > sizeof(protocol) - 1)
434 len = sizeof(protocol) - 1;
435 memcpy(protocol, filename, len);
436 protocol[len] = '\0';
437 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
438 if (drv1->protocol_name &&
439 !strcmp(drv1->protocol_name, protocol)) {
440 return drv1;
441 }
442 }
443 return NULL;
444 }
445
446 static int find_image_format(const char *filename, BlockDriver **pdrv)
447 {
448 int ret, score, score_max;
449 BlockDriver *drv1, *drv;
450 uint8_t buf[2048];
451 BlockDriverState *bs;
452
453 ret = bdrv_file_open(&bs, filename, 0);
454 if (ret < 0) {
455 *pdrv = NULL;
456 return ret;
457 }
458
459 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
460 if (bs->sg || !bdrv_is_inserted(bs)) {
461 bdrv_delete(bs);
462 drv = bdrv_find_format("raw");
463 if (!drv) {
464 ret = -ENOENT;
465 }
466 *pdrv = drv;
467 return ret;
468 }
469
470 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
471 bdrv_delete(bs);
472 if (ret < 0) {
473 *pdrv = NULL;
474 return ret;
475 }
476
477 score_max = 0;
478 drv = NULL;
479 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
480 if (drv1->bdrv_probe) {
481 score = drv1->bdrv_probe(buf, ret, filename);
482 if (score > score_max) {
483 score_max = score;
484 drv = drv1;
485 }
486 }
487 }
488 if (!drv) {
489 ret = -ENOENT;
490 }
491 *pdrv = drv;
492 return ret;
493 }
494
495 /**
496 * Set the current 'total_sectors' value
497 */
498 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
499 {
500 BlockDriver *drv = bs->drv;
501
502 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
503 if (bs->sg)
504 return 0;
505
506 /* query actual device if possible, otherwise just trust the hint */
507 if (drv->bdrv_getlength) {
508 int64_t length = drv->bdrv_getlength(bs);
509 if (length < 0) {
510 return length;
511 }
512 hint = length >> BDRV_SECTOR_BITS;
513 }
514
515 bs->total_sectors = hint;
516 return 0;
517 }
518
519 /**
520 * Set open flags for a given cache mode
521 *
522 * Return 0 on success, -1 if the cache mode was invalid.
523 */
524 int bdrv_parse_cache_flags(const char *mode, int *flags)
525 {
526 *flags &= ~BDRV_O_CACHE_MASK;
527
528 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
529 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
530 } else if (!strcmp(mode, "directsync")) {
531 *flags |= BDRV_O_NOCACHE;
532 } else if (!strcmp(mode, "writeback")) {
533 *flags |= BDRV_O_CACHE_WB;
534 } else if (!strcmp(mode, "unsafe")) {
535 *flags |= BDRV_O_CACHE_WB;
536 *flags |= BDRV_O_NO_FLUSH;
537 } else if (!strcmp(mode, "writethrough")) {
538 /* this is the default */
539 } else {
540 return -1;
541 }
542
543 return 0;
544 }
545
546 /**
547 * The copy-on-read flag is actually a reference count so multiple users may
548 * use the feature without worrying about clobbering its previous state.
549 * Copy-on-read stays enabled until all users have called to disable it.
550 */
551 void bdrv_enable_copy_on_read(BlockDriverState *bs)
552 {
553 bs->copy_on_read++;
554 }
555
556 void bdrv_disable_copy_on_read(BlockDriverState *bs)
557 {
558 assert(bs->copy_on_read > 0);
559 bs->copy_on_read--;
560 }
561
562 /*
563 * Common part for opening disk images and files
564 */
565 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
566 int flags, BlockDriver *drv)
567 {
568 int ret, open_flags;
569
570 assert(drv != NULL);
571
572 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
573
574 bs->file = NULL;
575 bs->total_sectors = 0;
576 bs->encrypted = 0;
577 bs->valid_key = 0;
578 bs->sg = 0;
579 bs->open_flags = flags;
580 bs->growable = 0;
581 bs->buffer_alignment = 512;
582
583 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
584 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
585 bdrv_enable_copy_on_read(bs);
586 }
587
588 pstrcpy(bs->filename, sizeof(bs->filename), filename);
589 bs->backing_file[0] = '\0';
590
591 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
592 return -ENOTSUP;
593 }
594
595 bs->drv = drv;
596 bs->opaque = g_malloc0(drv->instance_size);
597
598 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
599
600 /*
601 * Clear flags that are internal to the block layer before opening the
602 * image.
603 */
604 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
605
606 /*
607 * Snapshots should be writable.
608 */
609 if (bs->is_temporary) {
610 open_flags |= BDRV_O_RDWR;
611 }
612
613 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
614
615 /* Open the image, either directly or using a protocol */
616 if (drv->bdrv_file_open) {
617 ret = drv->bdrv_file_open(bs, filename, open_flags);
618 } else {
619 ret = bdrv_file_open(&bs->file, filename, open_flags);
620 if (ret >= 0) {
621 ret = drv->bdrv_open(bs, open_flags);
622 }
623 }
624
625 if (ret < 0) {
626 goto free_and_fail;
627 }
628
629 ret = refresh_total_sectors(bs, bs->total_sectors);
630 if (ret < 0) {
631 goto free_and_fail;
632 }
633
634 #ifndef _WIN32
635 if (bs->is_temporary) {
636 unlink(filename);
637 }
638 #endif
639 return 0;
640
641 free_and_fail:
642 if (bs->file) {
643 bdrv_delete(bs->file);
644 bs->file = NULL;
645 }
646 g_free(bs->opaque);
647 bs->opaque = NULL;
648 bs->drv = NULL;
649 return ret;
650 }
651
652 /*
653 * Opens a file using a protocol (file, host_device, nbd, ...)
654 */
655 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
656 {
657 BlockDriverState *bs;
658 BlockDriver *drv;
659 int ret;
660
661 drv = bdrv_find_protocol(filename);
662 if (!drv) {
663 return -ENOENT;
664 }
665
666 bs = bdrv_new("");
667 ret = bdrv_open_common(bs, filename, flags, drv);
668 if (ret < 0) {
669 bdrv_delete(bs);
670 return ret;
671 }
672 bs->growable = 1;
673 *pbs = bs;
674 return 0;
675 }
676
677 /*
678 * Opens a disk image (raw, qcow2, vmdk, ...)
679 */
680 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
681 BlockDriver *drv)
682 {
683 int ret;
684 char tmp_filename[PATH_MAX];
685
686 if (flags & BDRV_O_SNAPSHOT) {
687 BlockDriverState *bs1;
688 int64_t total_size;
689 int is_protocol = 0;
690 BlockDriver *bdrv_qcow2;
691 QEMUOptionParameter *options;
692 char backing_filename[PATH_MAX];
693
694 /* if snapshot, we create a temporary backing file and open it
695 instead of opening 'filename' directly */
696
697 /* if there is a backing file, use it */
698 bs1 = bdrv_new("");
699 ret = bdrv_open(bs1, filename, 0, drv);
700 if (ret < 0) {
701 bdrv_delete(bs1);
702 return ret;
703 }
704 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
705
706 if (bs1->drv && bs1->drv->protocol_name)
707 is_protocol = 1;
708
709 bdrv_delete(bs1);
710
711 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
712
713 /* Real path is meaningless for protocols */
714 if (is_protocol)
715 snprintf(backing_filename, sizeof(backing_filename),
716 "%s", filename);
717 else if (!realpath(filename, backing_filename))
718 return -errno;
719
720 bdrv_qcow2 = bdrv_find_format("qcow2");
721 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
722
723 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
724 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
725 if (drv) {
726 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
727 drv->format_name);
728 }
729
730 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
731 free_option_parameters(options);
732 if (ret < 0) {
733 return ret;
734 }
735
736 filename = tmp_filename;
737 drv = bdrv_qcow2;
738 bs->is_temporary = 1;
739 }
740
741 /* Find the right image format driver */
742 if (!drv) {
743 ret = find_image_format(filename, &drv);
744 }
745
746 if (!drv) {
747 goto unlink_and_fail;
748 }
749
750 /* Open the image */
751 ret = bdrv_open_common(bs, filename, flags, drv);
752 if (ret < 0) {
753 goto unlink_and_fail;
754 }
755
756 /* If there is a backing file, use it */
757 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
758 char backing_filename[PATH_MAX];
759 int back_flags;
760 BlockDriver *back_drv = NULL;
761
762 bs->backing_hd = bdrv_new("");
763
764 if (path_has_protocol(bs->backing_file)) {
765 pstrcpy(backing_filename, sizeof(backing_filename),
766 bs->backing_file);
767 } else {
768 path_combine(backing_filename, sizeof(backing_filename),
769 filename, bs->backing_file);
770 }
771
772 if (bs->backing_format[0] != '\0') {
773 back_drv = bdrv_find_format(bs->backing_format);
774 }
775
776 /* backing files always opened read-only */
777 back_flags =
778 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
779
780 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
781 if (ret < 0) {
782 bdrv_close(bs);
783 return ret;
784 }
785 if (bs->is_temporary) {
786 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
787 } else {
788 /* base image inherits from "parent" */
789 bs->backing_hd->keep_read_only = bs->keep_read_only;
790 }
791 }
792
793 if (!bdrv_key_required(bs)) {
794 bdrv_dev_change_media_cb(bs, true);
795 }
796
797 /* throttling disk I/O limits */
798 if (bs->io_limits_enabled) {
799 bdrv_io_limits_enable(bs);
800 }
801
802 return 0;
803
804 unlink_and_fail:
805 if (bs->is_temporary) {
806 unlink(filename);
807 }
808 return ret;
809 }
810
811 void bdrv_close(BlockDriverState *bs)
812 {
813 if (bs->drv) {
814 if (bs == bs_snapshots) {
815 bs_snapshots = NULL;
816 }
817 if (bs->backing_hd) {
818 bdrv_delete(bs->backing_hd);
819 bs->backing_hd = NULL;
820 }
821 bs->drv->bdrv_close(bs);
822 g_free(bs->opaque);
823 #ifdef _WIN32
824 if (bs->is_temporary) {
825 unlink(bs->filename);
826 }
827 #endif
828 bs->opaque = NULL;
829 bs->drv = NULL;
830 bs->copy_on_read = 0;
831
832 if (bs->file != NULL) {
833 bdrv_close(bs->file);
834 }
835
836 bdrv_dev_change_media_cb(bs, false);
837 }
838
839 /*throttling disk I/O limits*/
840 if (bs->io_limits_enabled) {
841 bdrv_io_limits_disable(bs);
842 }
843 }
844
845 void bdrv_close_all(void)
846 {
847 BlockDriverState *bs;
848
849 QTAILQ_FOREACH(bs, &bdrv_states, list) {
850 bdrv_close(bs);
851 }
852 }
853
854 /*
855 * Wait for pending requests to complete across all BlockDriverStates
856 *
857 * This function does not flush data to disk, use bdrv_flush_all() for that
858 * after calling this function.
859 */
860 void bdrv_drain_all(void)
861 {
862 BlockDriverState *bs;
863
864 qemu_aio_flush();
865
866 /* If requests are still pending there is a bug somewhere */
867 QTAILQ_FOREACH(bs, &bdrv_states, list) {
868 assert(QLIST_EMPTY(&bs->tracked_requests));
869 assert(qemu_co_queue_empty(&bs->throttled_reqs));
870 }
871 }
872
873 /* make a BlockDriverState anonymous by removing from bdrv_state list.
874 Also, NULL terminate the device_name to prevent double remove */
875 void bdrv_make_anon(BlockDriverState *bs)
876 {
877 if (bs->device_name[0] != '\0') {
878 QTAILQ_REMOVE(&bdrv_states, bs, list);
879 }
880 bs->device_name[0] = '\0';
881 }
882
883 void bdrv_delete(BlockDriverState *bs)
884 {
885 assert(!bs->dev);
886
887 /* remove from list, if necessary */
888 bdrv_make_anon(bs);
889
890 bdrv_close(bs);
891 if (bs->file != NULL) {
892 bdrv_delete(bs->file);
893 }
894
895 assert(bs != bs_snapshots);
896 g_free(bs);
897 }
898
899 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
900 /* TODO change to DeviceState *dev when all users are qdevified */
901 {
902 if (bs->dev) {
903 return -EBUSY;
904 }
905 bs->dev = dev;
906 bdrv_iostatus_reset(bs);
907 return 0;
908 }
909
910 /* TODO qdevified devices don't use this, remove when devices are qdevified */
911 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
912 {
913 if (bdrv_attach_dev(bs, dev) < 0) {
914 abort();
915 }
916 }
917
918 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
919 /* TODO change to DeviceState *dev when all users are qdevified */
920 {
921 assert(bs->dev == dev);
922 bs->dev = NULL;
923 bs->dev_ops = NULL;
924 bs->dev_opaque = NULL;
925 bs->buffer_alignment = 512;
926 }
927
928 /* TODO change to return DeviceState * when all users are qdevified */
929 void *bdrv_get_attached_dev(BlockDriverState *bs)
930 {
931 return bs->dev;
932 }
933
934 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
935 void *opaque)
936 {
937 bs->dev_ops = ops;
938 bs->dev_opaque = opaque;
939 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
940 bs_snapshots = NULL;
941 }
942 }
943
944 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
945 {
946 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
947 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
948 }
949 }
950
951 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
952 {
953 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
954 }
955
956 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
957 {
958 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
959 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
960 }
961 }
962
963 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
964 {
965 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
966 return bs->dev_ops->is_tray_open(bs->dev_opaque);
967 }
968 return false;
969 }
970
971 static void bdrv_dev_resize_cb(BlockDriverState *bs)
972 {
973 if (bs->dev_ops && bs->dev_ops->resize_cb) {
974 bs->dev_ops->resize_cb(bs->dev_opaque);
975 }
976 }
977
978 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
979 {
980 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
981 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
982 }
983 return false;
984 }
985
986 /*
987 * Run consistency checks on an image
988 *
989 * Returns 0 if the check could be completed (it doesn't mean that the image is
990 * free of errors) or -errno when an internal error occurred. The results of the
991 * check are stored in res.
992 */
993 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
994 {
995 if (bs->drv->bdrv_check == NULL) {
996 return -ENOTSUP;
997 }
998
999 memset(res, 0, sizeof(*res));
1000 return bs->drv->bdrv_check(bs, res);
1001 }
1002
1003 #define COMMIT_BUF_SECTORS 2048
1004
1005 /* commit COW file into the raw image */
1006 int bdrv_commit(BlockDriverState *bs)
1007 {
1008 BlockDriver *drv = bs->drv;
1009 BlockDriver *backing_drv;
1010 int64_t sector, total_sectors;
1011 int n, ro, open_flags;
1012 int ret = 0, rw_ret = 0;
1013 uint8_t *buf;
1014 char filename[1024];
1015 BlockDriverState *bs_rw, *bs_ro;
1016
1017 if (!drv)
1018 return -ENOMEDIUM;
1019
1020 if (!bs->backing_hd) {
1021 return -ENOTSUP;
1022 }
1023
1024 if (bs->backing_hd->keep_read_only) {
1025 return -EACCES;
1026 }
1027
1028 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1029 return -EBUSY;
1030 }
1031
1032 backing_drv = bs->backing_hd->drv;
1033 ro = bs->backing_hd->read_only;
1034 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1035 open_flags = bs->backing_hd->open_flags;
1036
1037 if (ro) {
1038 /* re-open as RW */
1039 bdrv_delete(bs->backing_hd);
1040 bs->backing_hd = NULL;
1041 bs_rw = bdrv_new("");
1042 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1043 backing_drv);
1044 if (rw_ret < 0) {
1045 bdrv_delete(bs_rw);
1046 /* try to re-open read-only */
1047 bs_ro = bdrv_new("");
1048 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1049 backing_drv);
1050 if (ret < 0) {
1051 bdrv_delete(bs_ro);
1052 /* drive not functional anymore */
1053 bs->drv = NULL;
1054 return ret;
1055 }
1056 bs->backing_hd = bs_ro;
1057 return rw_ret;
1058 }
1059 bs->backing_hd = bs_rw;
1060 }
1061
1062 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1063 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1064
1065 for (sector = 0; sector < total_sectors; sector += n) {
1066 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1067
1068 if (bdrv_read(bs, sector, buf, n) != 0) {
1069 ret = -EIO;
1070 goto ro_cleanup;
1071 }
1072
1073 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1074 ret = -EIO;
1075 goto ro_cleanup;
1076 }
1077 }
1078 }
1079
1080 if (drv->bdrv_make_empty) {
1081 ret = drv->bdrv_make_empty(bs);
1082 bdrv_flush(bs);
1083 }
1084
1085 /*
1086 * Make sure all data we wrote to the backing device is actually
1087 * stable on disk.
1088 */
1089 if (bs->backing_hd)
1090 bdrv_flush(bs->backing_hd);
1091
1092 ro_cleanup:
1093 g_free(buf);
1094
1095 if (ro) {
1096 /* re-open as RO */
1097 bdrv_delete(bs->backing_hd);
1098 bs->backing_hd = NULL;
1099 bs_ro = bdrv_new("");
1100 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1101 backing_drv);
1102 if (ret < 0) {
1103 bdrv_delete(bs_ro);
1104 /* drive not functional anymore */
1105 bs->drv = NULL;
1106 return ret;
1107 }
1108 bs->backing_hd = bs_ro;
1109 bs->backing_hd->keep_read_only = 0;
1110 }
1111
1112 return ret;
1113 }
1114
1115 void bdrv_commit_all(void)
1116 {
1117 BlockDriverState *bs;
1118
1119 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1120 bdrv_commit(bs);
1121 }
1122 }
1123
1124 struct BdrvTrackedRequest {
1125 BlockDriverState *bs;
1126 int64_t sector_num;
1127 int nb_sectors;
1128 bool is_write;
1129 QLIST_ENTRY(BdrvTrackedRequest) list;
1130 Coroutine *co; /* owner, used for deadlock detection */
1131 CoQueue wait_queue; /* coroutines blocked on this request */
1132 };
1133
1134 /**
1135 * Remove an active request from the tracked requests list
1136 *
1137 * This function should be called when a tracked request is completing.
1138 */
1139 static void tracked_request_end(BdrvTrackedRequest *req)
1140 {
1141 QLIST_REMOVE(req, list);
1142 qemu_co_queue_restart_all(&req->wait_queue);
1143 }
1144
1145 /**
1146 * Add an active request to the tracked requests list
1147 */
1148 static void tracked_request_begin(BdrvTrackedRequest *req,
1149 BlockDriverState *bs,
1150 int64_t sector_num,
1151 int nb_sectors, bool is_write)
1152 {
1153 *req = (BdrvTrackedRequest){
1154 .bs = bs,
1155 .sector_num = sector_num,
1156 .nb_sectors = nb_sectors,
1157 .is_write = is_write,
1158 .co = qemu_coroutine_self(),
1159 };
1160
1161 qemu_co_queue_init(&req->wait_queue);
1162
1163 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1164 }
1165
1166 /**
1167 * Round a region to cluster boundaries
1168 */
1169 static void round_to_clusters(BlockDriverState *bs,
1170 int64_t sector_num, int nb_sectors,
1171 int64_t *cluster_sector_num,
1172 int *cluster_nb_sectors)
1173 {
1174 BlockDriverInfo bdi;
1175
1176 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1177 *cluster_sector_num = sector_num;
1178 *cluster_nb_sectors = nb_sectors;
1179 } else {
1180 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1181 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1182 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1183 nb_sectors, c);
1184 }
1185 }
1186
1187 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1188 int64_t sector_num, int nb_sectors) {
1189 /* aaaa bbbb */
1190 if (sector_num >= req->sector_num + req->nb_sectors) {
1191 return false;
1192 }
1193 /* bbbb aaaa */
1194 if (req->sector_num >= sector_num + nb_sectors) {
1195 return false;
1196 }
1197 return true;
1198 }
1199
1200 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1201 int64_t sector_num, int nb_sectors)
1202 {
1203 BdrvTrackedRequest *req;
1204 int64_t cluster_sector_num;
1205 int cluster_nb_sectors;
1206 bool retry;
1207
1208 /* If we touch the same cluster it counts as an overlap. This guarantees
1209 * that allocating writes will be serialized and not race with each other
1210 * for the same cluster. For example, in copy-on-read it ensures that the
1211 * CoR read and write operations are atomic and guest writes cannot
1212 * interleave between them.
1213 */
1214 round_to_clusters(bs, sector_num, nb_sectors,
1215 &cluster_sector_num, &cluster_nb_sectors);
1216
1217 do {
1218 retry = false;
1219 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1220 if (tracked_request_overlaps(req, cluster_sector_num,
1221 cluster_nb_sectors)) {
1222 /* Hitting this means there was a reentrant request, for
1223 * example, a block driver issuing nested requests. This must
1224 * never happen since it means deadlock.
1225 */
1226 assert(qemu_coroutine_self() != req->co);
1227
1228 qemu_co_queue_wait(&req->wait_queue);
1229 retry = true;
1230 break;
1231 }
1232 }
1233 } while (retry);
1234 }
1235
1236 /*
1237 * Return values:
1238 * 0 - success
1239 * -EINVAL - backing format specified, but no file
1240 * -ENOSPC - can't update the backing file because no space is left in the
1241 * image file header
1242 * -ENOTSUP - format driver doesn't support changing the backing file
1243 */
1244 int bdrv_change_backing_file(BlockDriverState *bs,
1245 const char *backing_file, const char *backing_fmt)
1246 {
1247 BlockDriver *drv = bs->drv;
1248
1249 if (drv->bdrv_change_backing_file != NULL) {
1250 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1251 } else {
1252 return -ENOTSUP;
1253 }
1254 }
1255
1256 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1257 size_t size)
1258 {
1259 int64_t len;
1260
1261 if (!bdrv_is_inserted(bs))
1262 return -ENOMEDIUM;
1263
1264 if (bs->growable)
1265 return 0;
1266
1267 len = bdrv_getlength(bs);
1268
1269 if (offset < 0)
1270 return -EIO;
1271
1272 if ((offset > len) || (len - offset < size))
1273 return -EIO;
1274
1275 return 0;
1276 }
1277
1278 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1279 int nb_sectors)
1280 {
1281 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1282 nb_sectors * BDRV_SECTOR_SIZE);
1283 }
1284
1285 typedef struct RwCo {
1286 BlockDriverState *bs;
1287 int64_t sector_num;
1288 int nb_sectors;
1289 QEMUIOVector *qiov;
1290 bool is_write;
1291 int ret;
1292 } RwCo;
1293
1294 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1295 {
1296 RwCo *rwco = opaque;
1297
1298 if (!rwco->is_write) {
1299 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1300 rwco->nb_sectors, rwco->qiov, 0);
1301 } else {
1302 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1303 rwco->nb_sectors, rwco->qiov);
1304 }
1305 }
1306
1307 /*
1308 * Process a synchronous request using coroutines
1309 */
1310 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1311 int nb_sectors, bool is_write)
1312 {
1313 QEMUIOVector qiov;
1314 struct iovec iov = {
1315 .iov_base = (void *)buf,
1316 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1317 };
1318 Coroutine *co;
1319 RwCo rwco = {
1320 .bs = bs,
1321 .sector_num = sector_num,
1322 .nb_sectors = nb_sectors,
1323 .qiov = &qiov,
1324 .is_write = is_write,
1325 .ret = NOT_DONE,
1326 };
1327
1328 qemu_iovec_init_external(&qiov, &iov, 1);
1329
1330 if (qemu_in_coroutine()) {
1331 /* Fast-path if already in coroutine context */
1332 bdrv_rw_co_entry(&rwco);
1333 } else {
1334 co = qemu_coroutine_create(bdrv_rw_co_entry);
1335 qemu_coroutine_enter(co, &rwco);
1336 while (rwco.ret == NOT_DONE) {
1337 qemu_aio_wait();
1338 }
1339 }
1340 return rwco.ret;
1341 }
1342
1343 /* return < 0 if error. See bdrv_write() for the return codes */
1344 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1345 uint8_t *buf, int nb_sectors)
1346 {
1347 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1348 }
1349
1350 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1351 int nb_sectors, int dirty)
1352 {
1353 int64_t start, end;
1354 unsigned long val, idx, bit;
1355
1356 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1357 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1358
1359 for (; start <= end; start++) {
1360 idx = start / (sizeof(unsigned long) * 8);
1361 bit = start % (sizeof(unsigned long) * 8);
1362 val = bs->dirty_bitmap[idx];
1363 if (dirty) {
1364 if (!(val & (1UL << bit))) {
1365 bs->dirty_count++;
1366 val |= 1UL << bit;
1367 }
1368 } else {
1369 if (val & (1UL << bit)) {
1370 bs->dirty_count--;
1371 val &= ~(1UL << bit);
1372 }
1373 }
1374 bs->dirty_bitmap[idx] = val;
1375 }
1376 }
1377
1378 /* Return < 0 if error. Important errors are:
1379 -EIO generic I/O error (may happen for all errors)
1380 -ENOMEDIUM No media inserted.
1381 -EINVAL Invalid sector number or nb_sectors
1382 -EACCES Trying to write a read-only device
1383 */
1384 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1385 const uint8_t *buf, int nb_sectors)
1386 {
1387 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1388 }
1389
1390 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1391 void *buf, int count1)
1392 {
1393 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1394 int len, nb_sectors, count;
1395 int64_t sector_num;
1396 int ret;
1397
1398 count = count1;
1399 /* first read to align to sector start */
1400 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1401 if (len > count)
1402 len = count;
1403 sector_num = offset >> BDRV_SECTOR_BITS;
1404 if (len > 0) {
1405 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1406 return ret;
1407 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1408 count -= len;
1409 if (count == 0)
1410 return count1;
1411 sector_num++;
1412 buf += len;
1413 }
1414
1415 /* read the sectors "in place" */
1416 nb_sectors = count >> BDRV_SECTOR_BITS;
1417 if (nb_sectors > 0) {
1418 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1419 return ret;
1420 sector_num += nb_sectors;
1421 len = nb_sectors << BDRV_SECTOR_BITS;
1422 buf += len;
1423 count -= len;
1424 }
1425
1426 /* add data from the last sector */
1427 if (count > 0) {
1428 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1429 return ret;
1430 memcpy(buf, tmp_buf, count);
1431 }
1432 return count1;
1433 }
1434
1435 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1436 const void *buf, int count1)
1437 {
1438 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1439 int len, nb_sectors, count;
1440 int64_t sector_num;
1441 int ret;
1442
1443 count = count1;
1444 /* first write to align to sector start */
1445 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1446 if (len > count)
1447 len = count;
1448 sector_num = offset >> BDRV_SECTOR_BITS;
1449 if (len > 0) {
1450 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1451 return ret;
1452 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1453 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1454 return ret;
1455 count -= len;
1456 if (count == 0)
1457 return count1;
1458 sector_num++;
1459 buf += len;
1460 }
1461
1462 /* write the sectors "in place" */
1463 nb_sectors = count >> BDRV_SECTOR_BITS;
1464 if (nb_sectors > 0) {
1465 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1466 return ret;
1467 sector_num += nb_sectors;
1468 len = nb_sectors << BDRV_SECTOR_BITS;
1469 buf += len;
1470 count -= len;
1471 }
1472
1473 /* add data from the last sector */
1474 if (count > 0) {
1475 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1476 return ret;
1477 memcpy(tmp_buf, buf, count);
1478 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1479 return ret;
1480 }
1481 return count1;
1482 }
1483
1484 /*
1485 * Writes to the file and ensures that no writes are reordered across this
1486 * request (acts as a barrier)
1487 *
1488 * Returns 0 on success, -errno in error cases.
1489 */
1490 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1491 const void *buf, int count)
1492 {
1493 int ret;
1494
1495 ret = bdrv_pwrite(bs, offset, buf, count);
1496 if (ret < 0) {
1497 return ret;
1498 }
1499
1500 /* No flush needed for cache modes that use O_DSYNC */
1501 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1502 bdrv_flush(bs);
1503 }
1504
1505 return 0;
1506 }
1507
1508 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1509 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1510 {
1511 /* Perform I/O through a temporary buffer so that users who scribble over
1512 * their read buffer while the operation is in progress do not end up
1513 * modifying the image file. This is critical for zero-copy guest I/O
1514 * where anything might happen inside guest memory.
1515 */
1516 void *bounce_buffer;
1517
1518 struct iovec iov;
1519 QEMUIOVector bounce_qiov;
1520 int64_t cluster_sector_num;
1521 int cluster_nb_sectors;
1522 size_t skip_bytes;
1523 int ret;
1524
1525 /* Cover entire cluster so no additional backing file I/O is required when
1526 * allocating cluster in the image file.
1527 */
1528 round_to_clusters(bs, sector_num, nb_sectors,
1529 &cluster_sector_num, &cluster_nb_sectors);
1530
1531 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1532 cluster_sector_num, cluster_nb_sectors);
1533
1534 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1535 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1536 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1537
1538 ret = bs->drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1539 &bounce_qiov);
1540 if (ret < 0) {
1541 goto err;
1542 }
1543
1544 ret = bs->drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1545 &bounce_qiov);
1546 if (ret < 0) {
1547 /* It might be okay to ignore write errors for guest requests. If this
1548 * is a deliberate copy-on-read then we don't want to ignore the error.
1549 * Simply report it in all cases.
1550 */
1551 goto err;
1552 }
1553
1554 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1555 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1556 nb_sectors * BDRV_SECTOR_SIZE);
1557
1558 err:
1559 qemu_vfree(bounce_buffer);
1560 return ret;
1561 }
1562
1563 /*
1564 * Handle a read request in coroutine context
1565 */
1566 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1567 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1568 BdrvRequestFlags flags)
1569 {
1570 BlockDriver *drv = bs->drv;
1571 BdrvTrackedRequest req;
1572 int ret;
1573
1574 if (!drv) {
1575 return -ENOMEDIUM;
1576 }
1577 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1578 return -EIO;
1579 }
1580
1581 /* throttling disk read I/O */
1582 if (bs->io_limits_enabled) {
1583 bdrv_io_limits_intercept(bs, false, nb_sectors);
1584 }
1585
1586 if (bs->copy_on_read) {
1587 flags |= BDRV_REQ_COPY_ON_READ;
1588 }
1589 if (flags & BDRV_REQ_COPY_ON_READ) {
1590 bs->copy_on_read_in_flight++;
1591 }
1592
1593 if (bs->copy_on_read_in_flight) {
1594 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1595 }
1596
1597 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1598
1599 if (flags & BDRV_REQ_COPY_ON_READ) {
1600 int pnum;
1601
1602 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1603 if (ret < 0) {
1604 goto out;
1605 }
1606
1607 if (!ret || pnum != nb_sectors) {
1608 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1609 goto out;
1610 }
1611 }
1612
1613 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1614
1615 out:
1616 tracked_request_end(&req);
1617
1618 if (flags & BDRV_REQ_COPY_ON_READ) {
1619 bs->copy_on_read_in_flight--;
1620 }
1621
1622 return ret;
1623 }
1624
1625 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1626 int nb_sectors, QEMUIOVector *qiov)
1627 {
1628 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1629
1630 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1631 }
1632
1633 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1634 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1635 {
1636 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1637
1638 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1639 BDRV_REQ_COPY_ON_READ);
1640 }
1641
1642 /*
1643 * Handle a write request in coroutine context
1644 */
1645 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1646 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1647 {
1648 BlockDriver *drv = bs->drv;
1649 BdrvTrackedRequest req;
1650 int ret;
1651
1652 if (!bs->drv) {
1653 return -ENOMEDIUM;
1654 }
1655 if (bs->read_only) {
1656 return -EACCES;
1657 }
1658 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1659 return -EIO;
1660 }
1661
1662 /* throttling disk write I/O */
1663 if (bs->io_limits_enabled) {
1664 bdrv_io_limits_intercept(bs, true, nb_sectors);
1665 }
1666
1667 if (bs->copy_on_read_in_flight) {
1668 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1669 }
1670
1671 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1672
1673 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1674
1675 if (bs->dirty_bitmap) {
1676 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1677 }
1678
1679 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1680 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1681 }
1682
1683 tracked_request_end(&req);
1684
1685 return ret;
1686 }
1687
1688 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1689 int nb_sectors, QEMUIOVector *qiov)
1690 {
1691 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1692
1693 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov);
1694 }
1695
1696 /**
1697 * Truncate file to 'offset' bytes (needed only for file protocols)
1698 */
1699 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1700 {
1701 BlockDriver *drv = bs->drv;
1702 int ret;
1703 if (!drv)
1704 return -ENOMEDIUM;
1705 if (!drv->bdrv_truncate)
1706 return -ENOTSUP;
1707 if (bs->read_only)
1708 return -EACCES;
1709 if (bdrv_in_use(bs))
1710 return -EBUSY;
1711 ret = drv->bdrv_truncate(bs, offset);
1712 if (ret == 0) {
1713 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1714 bdrv_dev_resize_cb(bs);
1715 }
1716 return ret;
1717 }
1718
1719 /**
1720 * Length of a allocated file in bytes. Sparse files are counted by actual
1721 * allocated space. Return < 0 if error or unknown.
1722 */
1723 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1724 {
1725 BlockDriver *drv = bs->drv;
1726 if (!drv) {
1727 return -ENOMEDIUM;
1728 }
1729 if (drv->bdrv_get_allocated_file_size) {
1730 return drv->bdrv_get_allocated_file_size(bs);
1731 }
1732 if (bs->file) {
1733 return bdrv_get_allocated_file_size(bs->file);
1734 }
1735 return -ENOTSUP;
1736 }
1737
1738 /**
1739 * Length of a file in bytes. Return < 0 if error or unknown.
1740 */
1741 int64_t bdrv_getlength(BlockDriverState *bs)
1742 {
1743 BlockDriver *drv = bs->drv;
1744 if (!drv)
1745 return -ENOMEDIUM;
1746
1747 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
1748 if (drv->bdrv_getlength) {
1749 return drv->bdrv_getlength(bs);
1750 }
1751 }
1752 return bs->total_sectors * BDRV_SECTOR_SIZE;
1753 }
1754
1755 /* return 0 as number of sectors if no device present or error */
1756 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1757 {
1758 int64_t length;
1759 length = bdrv_getlength(bs);
1760 if (length < 0)
1761 length = 0;
1762 else
1763 length = length >> BDRV_SECTOR_BITS;
1764 *nb_sectors_ptr = length;
1765 }
1766
1767 struct partition {
1768 uint8_t boot_ind; /* 0x80 - active */
1769 uint8_t head; /* starting head */
1770 uint8_t sector; /* starting sector */
1771 uint8_t cyl; /* starting cylinder */
1772 uint8_t sys_ind; /* What partition type */
1773 uint8_t end_head; /* end head */
1774 uint8_t end_sector; /* end sector */
1775 uint8_t end_cyl; /* end cylinder */
1776 uint32_t start_sect; /* starting sector counting from 0 */
1777 uint32_t nr_sects; /* nr of sectors in partition */
1778 } QEMU_PACKED;
1779
1780 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1781 static int guess_disk_lchs(BlockDriverState *bs,
1782 int *pcylinders, int *pheads, int *psectors)
1783 {
1784 uint8_t buf[BDRV_SECTOR_SIZE];
1785 int ret, i, heads, sectors, cylinders;
1786 struct partition *p;
1787 uint32_t nr_sects;
1788 uint64_t nb_sectors;
1789
1790 bdrv_get_geometry(bs, &nb_sectors);
1791
1792 ret = bdrv_read(bs, 0, buf, 1);
1793 if (ret < 0)
1794 return -1;
1795 /* test msdos magic */
1796 if (buf[510] != 0x55 || buf[511] != 0xaa)
1797 return -1;
1798 for(i = 0; i < 4; i++) {
1799 p = ((struct partition *)(buf + 0x1be)) + i;
1800 nr_sects = le32_to_cpu(p->nr_sects);
1801 if (nr_sects && p->end_head) {
1802 /* We make the assumption that the partition terminates on
1803 a cylinder boundary */
1804 heads = p->end_head + 1;
1805 sectors = p->end_sector & 63;
1806 if (sectors == 0)
1807 continue;
1808 cylinders = nb_sectors / (heads * sectors);
1809 if (cylinders < 1 || cylinders > 16383)
1810 continue;
1811 *pheads = heads;
1812 *psectors = sectors;
1813 *pcylinders = cylinders;
1814 #if 0
1815 printf("guessed geometry: LCHS=%d %d %d\n",
1816 cylinders, heads, sectors);
1817 #endif
1818 return 0;
1819 }
1820 }
1821 return -1;
1822 }
1823
1824 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1825 {
1826 int translation, lba_detected = 0;
1827 int cylinders, heads, secs;
1828 uint64_t nb_sectors;
1829
1830 /* if a geometry hint is available, use it */
1831 bdrv_get_geometry(bs, &nb_sectors);
1832 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1833 translation = bdrv_get_translation_hint(bs);
1834 if (cylinders != 0) {
1835 *pcyls = cylinders;
1836 *pheads = heads;
1837 *psecs = secs;
1838 } else {
1839 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1840 if (heads > 16) {
1841 /* if heads > 16, it means that a BIOS LBA
1842 translation was active, so the default
1843 hardware geometry is OK */
1844 lba_detected = 1;
1845 goto default_geometry;
1846 } else {
1847 *pcyls = cylinders;
1848 *pheads = heads;
1849 *psecs = secs;
1850 /* disable any translation to be in sync with
1851 the logical geometry */
1852 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1853 bdrv_set_translation_hint(bs,
1854 BIOS_ATA_TRANSLATION_NONE);
1855 }
1856 }
1857 } else {
1858 default_geometry:
1859 /* if no geometry, use a standard physical disk geometry */
1860 cylinders = nb_sectors / (16 * 63);
1861
1862 if (cylinders > 16383)
1863 cylinders = 16383;
1864 else if (cylinders < 2)
1865 cylinders = 2;
1866 *pcyls = cylinders;
1867 *pheads = 16;
1868 *psecs = 63;
1869 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1870 if ((*pcyls * *pheads) <= 131072) {
1871 bdrv_set_translation_hint(bs,
1872 BIOS_ATA_TRANSLATION_LARGE);
1873 } else {
1874 bdrv_set_translation_hint(bs,
1875 BIOS_ATA_TRANSLATION_LBA);
1876 }
1877 }
1878 }
1879 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1880 }
1881 }
1882
1883 void bdrv_set_geometry_hint(BlockDriverState *bs,
1884 int cyls, int heads, int secs)
1885 {
1886 bs->cyls = cyls;
1887 bs->heads = heads;
1888 bs->secs = secs;
1889 }
1890
1891 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1892 {
1893 bs->translation = translation;
1894 }
1895
1896 void bdrv_get_geometry_hint(BlockDriverState *bs,
1897 int *pcyls, int *pheads, int *psecs)
1898 {
1899 *pcyls = bs->cyls;
1900 *pheads = bs->heads;
1901 *psecs = bs->secs;
1902 }
1903
1904 /* throttling disk io limits */
1905 void bdrv_set_io_limits(BlockDriverState *bs,
1906 BlockIOLimit *io_limits)
1907 {
1908 bs->io_limits = *io_limits;
1909 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
1910 }
1911
1912 /* Recognize floppy formats */
1913 typedef struct FDFormat {
1914 FDriveType drive;
1915 uint8_t last_sect;
1916 uint8_t max_track;
1917 uint8_t max_head;
1918 } FDFormat;
1919
1920 static const FDFormat fd_formats[] = {
1921 /* First entry is default format */
1922 /* 1.44 MB 3"1/2 floppy disks */
1923 { FDRIVE_DRV_144, 18, 80, 1, },
1924 { FDRIVE_DRV_144, 20, 80, 1, },
1925 { FDRIVE_DRV_144, 21, 80, 1, },
1926 { FDRIVE_DRV_144, 21, 82, 1, },
1927 { FDRIVE_DRV_144, 21, 83, 1, },
1928 { FDRIVE_DRV_144, 22, 80, 1, },
1929 { FDRIVE_DRV_144, 23, 80, 1, },
1930 { FDRIVE_DRV_144, 24, 80, 1, },
1931 /* 2.88 MB 3"1/2 floppy disks */
1932 { FDRIVE_DRV_288, 36, 80, 1, },
1933 { FDRIVE_DRV_288, 39, 80, 1, },
1934 { FDRIVE_DRV_288, 40, 80, 1, },
1935 { FDRIVE_DRV_288, 44, 80, 1, },
1936 { FDRIVE_DRV_288, 48, 80, 1, },
1937 /* 720 kB 3"1/2 floppy disks */
1938 { FDRIVE_DRV_144, 9, 80, 1, },
1939 { FDRIVE_DRV_144, 10, 80, 1, },
1940 { FDRIVE_DRV_144, 10, 82, 1, },
1941 { FDRIVE_DRV_144, 10, 83, 1, },
1942 { FDRIVE_DRV_144, 13, 80, 1, },
1943 { FDRIVE_DRV_144, 14, 80, 1, },
1944 /* 1.2 MB 5"1/4 floppy disks */
1945 { FDRIVE_DRV_120, 15, 80, 1, },
1946 { FDRIVE_DRV_120, 18, 80, 1, },
1947 { FDRIVE_DRV_120, 18, 82, 1, },
1948 { FDRIVE_DRV_120, 18, 83, 1, },
1949 { FDRIVE_DRV_120, 20, 80, 1, },
1950 /* 720 kB 5"1/4 floppy disks */
1951 { FDRIVE_DRV_120, 9, 80, 1, },
1952 { FDRIVE_DRV_120, 11, 80, 1, },
1953 /* 360 kB 5"1/4 floppy disks */
1954 { FDRIVE_DRV_120, 9, 40, 1, },
1955 { FDRIVE_DRV_120, 9, 40, 0, },
1956 { FDRIVE_DRV_120, 10, 41, 1, },
1957 { FDRIVE_DRV_120, 10, 42, 1, },
1958 /* 320 kB 5"1/4 floppy disks */
1959 { FDRIVE_DRV_120, 8, 40, 1, },
1960 { FDRIVE_DRV_120, 8, 40, 0, },
1961 /* 360 kB must match 5"1/4 better than 3"1/2... */
1962 { FDRIVE_DRV_144, 9, 80, 0, },
1963 /* end */
1964 { FDRIVE_DRV_NONE, -1, -1, 0, },
1965 };
1966
1967 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
1968 int *max_track, int *last_sect,
1969 FDriveType drive_in, FDriveType *drive)
1970 {
1971 const FDFormat *parse;
1972 uint64_t nb_sectors, size;
1973 int i, first_match, match;
1974
1975 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
1976 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
1977 /* User defined disk */
1978 } else {
1979 bdrv_get_geometry(bs, &nb_sectors);
1980 match = -1;
1981 first_match = -1;
1982 for (i = 0; ; i++) {
1983 parse = &fd_formats[i];
1984 if (parse->drive == FDRIVE_DRV_NONE) {
1985 break;
1986 }
1987 if (drive_in == parse->drive ||
1988 drive_in == FDRIVE_DRV_NONE) {
1989 size = (parse->max_head + 1) * parse->max_track *
1990 parse->last_sect;
1991 if (nb_sectors == size) {
1992 match = i;
1993 break;
1994 }
1995 if (first_match == -1) {
1996 first_match = i;
1997 }
1998 }
1999 }
2000 if (match == -1) {
2001 if (first_match == -1) {
2002 match = 1;
2003 } else {
2004 match = first_match;
2005 }
2006 parse = &fd_formats[match];
2007 }
2008 *nb_heads = parse->max_head + 1;
2009 *max_track = parse->max_track;
2010 *last_sect = parse->last_sect;
2011 *drive = parse->drive;
2012 }
2013 }
2014
2015 int bdrv_get_translation_hint(BlockDriverState *bs)
2016 {
2017 return bs->translation;
2018 }
2019
2020 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2021 BlockErrorAction on_write_error)
2022 {
2023 bs->on_read_error = on_read_error;
2024 bs->on_write_error = on_write_error;
2025 }
2026
2027 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2028 {
2029 return is_read ? bs->on_read_error : bs->on_write_error;
2030 }
2031
2032 int bdrv_is_read_only(BlockDriverState *bs)
2033 {
2034 return bs->read_only;
2035 }
2036
2037 int bdrv_is_sg(BlockDriverState *bs)
2038 {
2039 return bs->sg;
2040 }
2041
2042 int bdrv_enable_write_cache(BlockDriverState *bs)
2043 {
2044 return bs->enable_write_cache;
2045 }
2046
2047 int bdrv_is_encrypted(BlockDriverState *bs)
2048 {
2049 if (bs->backing_hd && bs->backing_hd->encrypted)
2050 return 1;
2051 return bs->encrypted;
2052 }
2053
2054 int bdrv_key_required(BlockDriverState *bs)
2055 {
2056 BlockDriverState *backing_hd = bs->backing_hd;
2057
2058 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2059 return 1;
2060 return (bs->encrypted && !bs->valid_key);
2061 }
2062
2063 int bdrv_set_key(BlockDriverState *bs, const char *key)
2064 {
2065 int ret;
2066 if (bs->backing_hd && bs->backing_hd->encrypted) {
2067 ret = bdrv_set_key(bs->backing_hd, key);
2068 if (ret < 0)
2069 return ret;
2070 if (!bs->encrypted)
2071 return 0;
2072 }
2073 if (!bs->encrypted) {
2074 return -EINVAL;
2075 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2076 return -ENOMEDIUM;
2077 }
2078 ret = bs->drv->bdrv_set_key(bs, key);
2079 if (ret < 0) {
2080 bs->valid_key = 0;
2081 } else if (!bs->valid_key) {
2082 bs->valid_key = 1;
2083 /* call the change callback now, we skipped it on open */
2084 bdrv_dev_change_media_cb(bs, true);
2085 }
2086 return ret;
2087 }
2088
2089 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2090 {
2091 if (!bs->drv) {
2092 buf[0] = '\0';
2093 } else {
2094 pstrcpy(buf, buf_size, bs->drv->format_name);
2095 }
2096 }
2097
2098 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2099 void *opaque)
2100 {
2101 BlockDriver *drv;
2102
2103 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2104 it(opaque, drv->format_name);
2105 }
2106 }
2107
2108 BlockDriverState *bdrv_find(const char *name)
2109 {
2110 BlockDriverState *bs;
2111
2112 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2113 if (!strcmp(name, bs->device_name)) {
2114 return bs;
2115 }
2116 }
2117 return NULL;
2118 }
2119
2120 BlockDriverState *bdrv_next(BlockDriverState *bs)
2121 {
2122 if (!bs) {
2123 return QTAILQ_FIRST(&bdrv_states);
2124 }
2125 return QTAILQ_NEXT(bs, list);
2126 }
2127
2128 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2129 {
2130 BlockDriverState *bs;
2131
2132 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2133 it(opaque, bs);
2134 }
2135 }
2136
2137 const char *bdrv_get_device_name(BlockDriverState *bs)
2138 {
2139 return bs->device_name;
2140 }
2141
2142 void bdrv_flush_all(void)
2143 {
2144 BlockDriverState *bs;
2145
2146 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2147 if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
2148 bdrv_flush(bs);
2149 }
2150 }
2151 }
2152
2153 int bdrv_has_zero_init(BlockDriverState *bs)
2154 {
2155 assert(bs->drv);
2156
2157 if (bs->drv->bdrv_has_zero_init) {
2158 return bs->drv->bdrv_has_zero_init(bs);
2159 }
2160
2161 return 1;
2162 }
2163
2164 typedef struct BdrvCoIsAllocatedData {
2165 BlockDriverState *bs;
2166 int64_t sector_num;
2167 int nb_sectors;
2168 int *pnum;
2169 int ret;
2170 bool done;
2171 } BdrvCoIsAllocatedData;
2172
2173 /*
2174 * Returns true iff the specified sector is present in the disk image. Drivers
2175 * not implementing the functionality are assumed to not support backing files,
2176 * hence all their sectors are reported as allocated.
2177 *
2178 * If 'sector_num' is beyond the end of the disk image the return value is 0
2179 * and 'pnum' is set to 0.
2180 *
2181 * 'pnum' is set to the number of sectors (including and immediately following
2182 * the specified sector) that are known to be in the same
2183 * allocated/unallocated state.
2184 *
2185 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2186 * beyond the end of the disk image it will be clamped.
2187 */
2188 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2189 int nb_sectors, int *pnum)
2190 {
2191 int64_t n;
2192
2193 if (sector_num >= bs->total_sectors) {
2194 *pnum = 0;
2195 return 0;
2196 }
2197
2198 n = bs->total_sectors - sector_num;
2199 if (n < nb_sectors) {
2200 nb_sectors = n;
2201 }
2202
2203 if (!bs->drv->bdrv_co_is_allocated) {
2204 *pnum = nb_sectors;
2205 return 1;
2206 }
2207
2208 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2209 }
2210
2211 /* Coroutine wrapper for bdrv_is_allocated() */
2212 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2213 {
2214 BdrvCoIsAllocatedData *data = opaque;
2215 BlockDriverState *bs = data->bs;
2216
2217 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2218 data->pnum);
2219 data->done = true;
2220 }
2221
2222 /*
2223 * Synchronous wrapper around bdrv_co_is_allocated().
2224 *
2225 * See bdrv_co_is_allocated() for details.
2226 */
2227 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2228 int *pnum)
2229 {
2230 Coroutine *co;
2231 BdrvCoIsAllocatedData data = {
2232 .bs = bs,
2233 .sector_num = sector_num,
2234 .nb_sectors = nb_sectors,
2235 .pnum = pnum,
2236 .done = false,
2237 };
2238
2239 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2240 qemu_coroutine_enter(co, &data);
2241 while (!data.done) {
2242 qemu_aio_wait();
2243 }
2244 return data.ret;
2245 }
2246
2247 void bdrv_mon_event(const BlockDriverState *bdrv,
2248 BlockMonEventAction action, int is_read)
2249 {
2250 QObject *data;
2251 const char *action_str;
2252
2253 switch (action) {
2254 case BDRV_ACTION_REPORT:
2255 action_str = "report";
2256 break;
2257 case BDRV_ACTION_IGNORE:
2258 action_str = "ignore";
2259 break;
2260 case BDRV_ACTION_STOP:
2261 action_str = "stop";
2262 break;
2263 default:
2264 abort();
2265 }
2266
2267 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
2268 bdrv->device_name,
2269 action_str,
2270 is_read ? "read" : "write");
2271 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
2272
2273 qobject_decref(data);
2274 }
2275
2276 BlockInfoList *qmp_query_block(Error **errp)
2277 {
2278 BlockInfoList *head = NULL, *cur_item = NULL;
2279 BlockDriverState *bs;
2280
2281 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2282 BlockInfoList *info = g_malloc0(sizeof(*info));
2283
2284 info->value = g_malloc0(sizeof(*info->value));
2285 info->value->device = g_strdup(bs->device_name);
2286 info->value->type = g_strdup("unknown");
2287 info->value->locked = bdrv_dev_is_medium_locked(bs);
2288 info->value->removable = bdrv_dev_has_removable_media(bs);
2289
2290 if (bdrv_dev_has_removable_media(bs)) {
2291 info->value->has_tray_open = true;
2292 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2293 }
2294
2295 if (bdrv_iostatus_is_enabled(bs)) {
2296 info->value->has_io_status = true;
2297 info->value->io_status = bs->iostatus;
2298 }
2299
2300 if (bs->drv) {
2301 info->value->has_inserted = true;
2302 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2303 info->value->inserted->file = g_strdup(bs->filename);
2304 info->value->inserted->ro = bs->read_only;
2305 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2306 info->value->inserted->encrypted = bs->encrypted;
2307 if (bs->backing_file[0]) {
2308 info->value->inserted->has_backing_file = true;
2309 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2310 }
2311
2312 if (bs->io_limits_enabled) {
2313 info->value->inserted->bps =
2314 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2315 info->value->inserted->bps_rd =
2316 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2317 info->value->inserted->bps_wr =
2318 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2319 info->value->inserted->iops =
2320 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2321 info->value->inserted->iops_rd =
2322 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2323 info->value->inserted->iops_wr =
2324 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2325 }
2326 }
2327
2328 /* XXX: waiting for the qapi to support GSList */
2329 if (!cur_item) {
2330 head = cur_item = info;
2331 } else {
2332 cur_item->next = info;
2333 cur_item = info;
2334 }
2335 }
2336
2337 return head;
2338 }
2339
2340 /* Consider exposing this as a full fledged QMP command */
2341 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2342 {
2343 BlockStats *s;
2344
2345 s = g_malloc0(sizeof(*s));
2346
2347 if (bs->device_name[0]) {
2348 s->has_device = true;
2349 s->device = g_strdup(bs->device_name);
2350 }
2351
2352 s->stats = g_malloc0(sizeof(*s->stats));
2353 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2354 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2355 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2356 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2357 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2358 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2359 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2360 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2361 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2362
2363 if (bs->file) {
2364 s->has_parent = true;
2365 s->parent = qmp_query_blockstat(bs->file, NULL);
2366 }
2367
2368 return s;
2369 }
2370
2371 BlockStatsList *qmp_query_blockstats(Error **errp)
2372 {
2373 BlockStatsList *head = NULL, *cur_item = NULL;
2374 BlockDriverState *bs;
2375
2376 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2377 BlockStatsList *info = g_malloc0(sizeof(*info));
2378 info->value = qmp_query_blockstat(bs, NULL);
2379
2380 /* XXX: waiting for the qapi to support GSList */
2381 if (!cur_item) {
2382 head = cur_item = info;
2383 } else {
2384 cur_item->next = info;
2385 cur_item = info;
2386 }
2387 }
2388
2389 return head;
2390 }
2391
2392 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2393 {
2394 if (bs->backing_hd && bs->backing_hd->encrypted)
2395 return bs->backing_file;
2396 else if (bs->encrypted)
2397 return bs->filename;
2398 else
2399 return NULL;
2400 }
2401
2402 void bdrv_get_backing_filename(BlockDriverState *bs,
2403 char *filename, int filename_size)
2404 {
2405 pstrcpy(filename, filename_size, bs->backing_file);
2406 }
2407
2408 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2409 const uint8_t *buf, int nb_sectors)
2410 {
2411 BlockDriver *drv = bs->drv;
2412 if (!drv)
2413 return -ENOMEDIUM;
2414 if (!drv->bdrv_write_compressed)
2415 return -ENOTSUP;
2416 if (bdrv_check_request(bs, sector_num, nb_sectors))
2417 return -EIO;
2418
2419 if (bs->dirty_bitmap) {
2420 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2421 }
2422
2423 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2424 }
2425
2426 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2427 {
2428 BlockDriver *drv = bs->drv;
2429 if (!drv)
2430 return -ENOMEDIUM;
2431 if (!drv->bdrv_get_info)
2432 return -ENOTSUP;
2433 memset(bdi, 0, sizeof(*bdi));
2434 return drv->bdrv_get_info(bs, bdi);
2435 }
2436
2437 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2438 int64_t pos, int size)
2439 {
2440 BlockDriver *drv = bs->drv;
2441 if (!drv)
2442 return -ENOMEDIUM;
2443 if (drv->bdrv_save_vmstate)
2444 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2445 if (bs->file)
2446 return bdrv_save_vmstate(bs->file, buf, pos, size);
2447 return -ENOTSUP;
2448 }
2449
2450 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2451 int64_t pos, int size)
2452 {
2453 BlockDriver *drv = bs->drv;
2454 if (!drv)
2455 return -ENOMEDIUM;
2456 if (drv->bdrv_load_vmstate)
2457 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2458 if (bs->file)
2459 return bdrv_load_vmstate(bs->file, buf, pos, size);
2460 return -ENOTSUP;
2461 }
2462
2463 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2464 {
2465 BlockDriver *drv = bs->drv;
2466
2467 if (!drv || !drv->bdrv_debug_event) {
2468 return;
2469 }
2470
2471 return drv->bdrv_debug_event(bs, event);
2472
2473 }
2474
2475 /**************************************************************/
2476 /* handling of snapshots */
2477
2478 int bdrv_can_snapshot(BlockDriverState *bs)
2479 {
2480 BlockDriver *drv = bs->drv;
2481 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2482 return 0;
2483 }
2484
2485 if (!drv->bdrv_snapshot_create) {
2486 if (bs->file != NULL) {
2487 return bdrv_can_snapshot(bs->file);
2488 }
2489 return 0;
2490 }
2491
2492 return 1;
2493 }
2494
2495 int bdrv_is_snapshot(BlockDriverState *bs)
2496 {
2497 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2498 }
2499
2500 BlockDriverState *bdrv_snapshots(void)
2501 {
2502 BlockDriverState *bs;
2503
2504 if (bs_snapshots) {
2505 return bs_snapshots;
2506 }
2507
2508 bs = NULL;
2509 while ((bs = bdrv_next(bs))) {
2510 if (bdrv_can_snapshot(bs)) {
2511 bs_snapshots = bs;
2512 return bs;
2513 }
2514 }
2515 return NULL;
2516 }
2517
2518 int bdrv_snapshot_create(BlockDriverState *bs,
2519 QEMUSnapshotInfo *sn_info)
2520 {
2521 BlockDriver *drv = bs->drv;
2522 if (!drv)
2523 return -ENOMEDIUM;
2524 if (drv->bdrv_snapshot_create)
2525 return drv->bdrv_snapshot_create(bs, sn_info);
2526 if (bs->file)
2527 return bdrv_snapshot_create(bs->file, sn_info);
2528 return -ENOTSUP;
2529 }
2530
2531 int bdrv_snapshot_goto(BlockDriverState *bs,
2532 const char *snapshot_id)
2533 {
2534 BlockDriver *drv = bs->drv;
2535 int ret, open_ret;
2536
2537 if (!drv)
2538 return -ENOMEDIUM;
2539 if (drv->bdrv_snapshot_goto)
2540 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2541
2542 if (bs->file) {
2543 drv->bdrv_close(bs);
2544 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2545 open_ret = drv->bdrv_open(bs, bs->open_flags);
2546 if (open_ret < 0) {
2547 bdrv_delete(bs->file);
2548 bs->drv = NULL;
2549 return open_ret;
2550 }
2551 return ret;
2552 }
2553
2554 return -ENOTSUP;
2555 }
2556
2557 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2558 {
2559 BlockDriver *drv = bs->drv;
2560 if (!drv)
2561 return -ENOMEDIUM;
2562 if (drv->bdrv_snapshot_delete)
2563 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2564 if (bs->file)
2565 return bdrv_snapshot_delete(bs->file, snapshot_id);
2566 return -ENOTSUP;
2567 }
2568
2569 int bdrv_snapshot_list(BlockDriverState *bs,
2570 QEMUSnapshotInfo **psn_info)
2571 {
2572 BlockDriver *drv = bs->drv;
2573 if (!drv)
2574 return -ENOMEDIUM;
2575 if (drv->bdrv_snapshot_list)
2576 return drv->bdrv_snapshot_list(bs, psn_info);
2577 if (bs->file)
2578 return bdrv_snapshot_list(bs->file, psn_info);
2579 return -ENOTSUP;
2580 }
2581
2582 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2583 const char *snapshot_name)
2584 {
2585 BlockDriver *drv = bs->drv;
2586 if (!drv) {
2587 return -ENOMEDIUM;
2588 }
2589 if (!bs->read_only) {
2590 return -EINVAL;
2591 }
2592 if (drv->bdrv_snapshot_load_tmp) {
2593 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2594 }
2595 return -ENOTSUP;
2596 }
2597
2598 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2599 const char *backing_file)
2600 {
2601 if (!bs->drv) {
2602 return NULL;
2603 }
2604
2605 if (bs->backing_hd) {
2606 if (strcmp(bs->backing_file, backing_file) == 0) {
2607 return bs->backing_hd;
2608 } else {
2609 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2610 }
2611 }
2612
2613 return NULL;
2614 }
2615
2616 #define NB_SUFFIXES 4
2617
2618 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2619 {
2620 static const char suffixes[NB_SUFFIXES] = "KMGT";
2621 int64_t base;
2622 int i;
2623
2624 if (size <= 999) {
2625 snprintf(buf, buf_size, "%" PRId64, size);
2626 } else {
2627 base = 1024;
2628 for(i = 0; i < NB_SUFFIXES; i++) {
2629 if (size < (10 * base)) {
2630 snprintf(buf, buf_size, "%0.1f%c",
2631 (double)size / base,
2632 suffixes[i]);
2633 break;
2634 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2635 snprintf(buf, buf_size, "%" PRId64 "%c",
2636 ((size + (base >> 1)) / base),
2637 suffixes[i]);
2638 break;
2639 }
2640 base = base * 1024;
2641 }
2642 }
2643 return buf;
2644 }
2645
2646 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2647 {
2648 char buf1[128], date_buf[128], clock_buf[128];
2649 #ifdef _WIN32
2650 struct tm *ptm;
2651 #else
2652 struct tm tm;
2653 #endif
2654 time_t ti;
2655 int64_t secs;
2656
2657 if (!sn) {
2658 snprintf(buf, buf_size,
2659 "%-10s%-20s%7s%20s%15s",
2660 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2661 } else {
2662 ti = sn->date_sec;
2663 #ifdef _WIN32
2664 ptm = localtime(&ti);
2665 strftime(date_buf, sizeof(date_buf),
2666 "%Y-%m-%d %H:%M:%S", ptm);
2667 #else
2668 localtime_r(&ti, &tm);
2669 strftime(date_buf, sizeof(date_buf),
2670 "%Y-%m-%d %H:%M:%S", &tm);
2671 #endif
2672 secs = sn->vm_clock_nsec / 1000000000;
2673 snprintf(clock_buf, sizeof(clock_buf),
2674 "%02d:%02d:%02d.%03d",
2675 (int)(secs / 3600),
2676 (int)((secs / 60) % 60),
2677 (int)(secs % 60),
2678 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2679 snprintf(buf, buf_size,
2680 "%-10s%-20s%7s%20s%15s",
2681 sn->id_str, sn->name,
2682 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2683 date_buf,
2684 clock_buf);
2685 }
2686 return buf;
2687 }
2688
2689 /**************************************************************/
2690 /* async I/Os */
2691
2692 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2693 QEMUIOVector *qiov, int nb_sectors,
2694 BlockDriverCompletionFunc *cb, void *opaque)
2695 {
2696 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2697
2698 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2699 cb, opaque, false);
2700 }
2701
2702 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2703 QEMUIOVector *qiov, int nb_sectors,
2704 BlockDriverCompletionFunc *cb, void *opaque)
2705 {
2706 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2707
2708 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2709 cb, opaque, true);
2710 }
2711
2712
2713 typedef struct MultiwriteCB {
2714 int error;
2715 int num_requests;
2716 int num_callbacks;
2717 struct {
2718 BlockDriverCompletionFunc *cb;
2719 void *opaque;
2720 QEMUIOVector *free_qiov;
2721 void *free_buf;
2722 } callbacks[];
2723 } MultiwriteCB;
2724
2725 static void multiwrite_user_cb(MultiwriteCB *mcb)
2726 {
2727 int i;
2728
2729 for (i = 0; i < mcb->num_callbacks; i++) {
2730 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2731 if (mcb->callbacks[i].free_qiov) {
2732 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2733 }
2734 g_free(mcb->callbacks[i].free_qiov);
2735 qemu_vfree(mcb->callbacks[i].free_buf);
2736 }
2737 }
2738
2739 static void multiwrite_cb(void *opaque, int ret)
2740 {
2741 MultiwriteCB *mcb = opaque;
2742
2743 trace_multiwrite_cb(mcb, ret);
2744
2745 if (ret < 0 && !mcb->error) {
2746 mcb->error = ret;
2747 }
2748
2749 mcb->num_requests--;
2750 if (mcb->num_requests == 0) {
2751 multiwrite_user_cb(mcb);
2752 g_free(mcb);
2753 }
2754 }
2755
2756 static int multiwrite_req_compare(const void *a, const void *b)
2757 {
2758 const BlockRequest *req1 = a, *req2 = b;
2759
2760 /*
2761 * Note that we can't simply subtract req2->sector from req1->sector
2762 * here as that could overflow the return value.
2763 */
2764 if (req1->sector > req2->sector) {
2765 return 1;
2766 } else if (req1->sector < req2->sector) {
2767 return -1;
2768 } else {
2769 return 0;
2770 }
2771 }
2772
2773 /*
2774 * Takes a bunch of requests and tries to merge them. Returns the number of
2775 * requests that remain after merging.
2776 */
2777 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2778 int num_reqs, MultiwriteCB *mcb)
2779 {
2780 int i, outidx;
2781
2782 // Sort requests by start sector
2783 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2784
2785 // Check if adjacent requests touch the same clusters. If so, combine them,
2786 // filling up gaps with zero sectors.
2787 outidx = 0;
2788 for (i = 1; i < num_reqs; i++) {
2789 int merge = 0;
2790 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2791
2792 // This handles the cases that are valid for all block drivers, namely
2793 // exactly sequential writes and overlapping writes.
2794 if (reqs[i].sector <= oldreq_last) {
2795 merge = 1;
2796 }
2797
2798 // The block driver may decide that it makes sense to combine requests
2799 // even if there is a gap of some sectors between them. In this case,
2800 // the gap is filled with zeros (therefore only applicable for yet
2801 // unused space in format like qcow2).
2802 if (!merge && bs->drv->bdrv_merge_requests) {
2803 merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2804 }
2805
2806 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2807 merge = 0;
2808 }
2809
2810 if (merge) {
2811 size_t size;
2812 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
2813 qemu_iovec_init(qiov,
2814 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2815
2816 // Add the first request to the merged one. If the requests are
2817 // overlapping, drop the last sectors of the first request.
2818 size = (reqs[i].sector - reqs[outidx].sector) << 9;
2819 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2820
2821 // We might need to add some zeros between the two requests
2822 if (reqs[i].sector > oldreq_last) {
2823 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2824 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2825 memset(buf, 0, zero_bytes);
2826 qemu_iovec_add(qiov, buf, zero_bytes);
2827 mcb->callbacks[i].free_buf = buf;
2828 }
2829
2830 // Add the second request
2831 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2832
2833 reqs[outidx].nb_sectors = qiov->size >> 9;
2834 reqs[outidx].qiov = qiov;
2835
2836 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2837 } else {
2838 outidx++;
2839 reqs[outidx].sector = reqs[i].sector;
2840 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2841 reqs[outidx].qiov = reqs[i].qiov;
2842 }
2843 }
2844
2845 return outidx + 1;
2846 }
2847
2848 /*
2849 * Submit multiple AIO write requests at once.
2850 *
2851 * On success, the function returns 0 and all requests in the reqs array have
2852 * been submitted. In error case this function returns -1, and any of the
2853 * requests may or may not be submitted yet. In particular, this means that the
2854 * callback will be called for some of the requests, for others it won't. The
2855 * caller must check the error field of the BlockRequest to wait for the right
2856 * callbacks (if error != 0, no callback will be called).
2857 *
2858 * The implementation may modify the contents of the reqs array, e.g. to merge
2859 * requests. However, the fields opaque and error are left unmodified as they
2860 * are used to signal failure for a single request to the caller.
2861 */
2862 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2863 {
2864 MultiwriteCB *mcb;
2865 int i;
2866
2867 /* don't submit writes if we don't have a medium */
2868 if (bs->drv == NULL) {
2869 for (i = 0; i < num_reqs; i++) {
2870 reqs[i].error = -ENOMEDIUM;
2871 }
2872 return -1;
2873 }
2874
2875 if (num_reqs == 0) {
2876 return 0;
2877 }
2878
2879 // Create MultiwriteCB structure
2880 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2881 mcb->num_requests = 0;
2882 mcb->num_callbacks = num_reqs;
2883
2884 for (i = 0; i < num_reqs; i++) {
2885 mcb->callbacks[i].cb = reqs[i].cb;
2886 mcb->callbacks[i].opaque = reqs[i].opaque;
2887 }
2888
2889 // Check for mergable requests
2890 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2891
2892 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
2893
2894 /* Run the aio requests. */
2895 mcb->num_requests = num_reqs;
2896 for (i = 0; i < num_reqs; i++) {
2897 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2898 reqs[i].nb_sectors, multiwrite_cb, mcb);
2899 }
2900
2901 return 0;
2902 }
2903
2904 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2905 {
2906 acb->pool->cancel(acb);
2907 }
2908
2909 /* block I/O throttling */
2910 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
2911 bool is_write, double elapsed_time, uint64_t *wait)
2912 {
2913 uint64_t bps_limit = 0;
2914 double bytes_limit, bytes_base, bytes_res;
2915 double slice_time, wait_time;
2916
2917 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
2918 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2919 } else if (bs->io_limits.bps[is_write]) {
2920 bps_limit = bs->io_limits.bps[is_write];
2921 } else {
2922 if (wait) {
2923 *wait = 0;
2924 }
2925
2926 return false;
2927 }
2928
2929 slice_time = bs->slice_end - bs->slice_start;
2930 slice_time /= (NANOSECONDS_PER_SECOND);
2931 bytes_limit = bps_limit * slice_time;
2932 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
2933 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
2934 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
2935 }
2936
2937 /* bytes_base: the bytes of data which have been read/written; and
2938 * it is obtained from the history statistic info.
2939 * bytes_res: the remaining bytes of data which need to be read/written.
2940 * (bytes_base + bytes_res) / bps_limit: used to calcuate
2941 * the total time for completing reading/writting all data.
2942 */
2943 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2944
2945 if (bytes_base + bytes_res <= bytes_limit) {
2946 if (wait) {
2947 *wait = 0;
2948 }
2949
2950 return false;
2951 }
2952
2953 /* Calc approx time to dispatch */
2954 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
2955
2956 /* When the I/O rate at runtime exceeds the limits,
2957 * bs->slice_end need to be extended in order that the current statistic
2958 * info can be kept until the timer fire, so it is increased and tuned
2959 * based on the result of experiment.
2960 */
2961 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
2962 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
2963 if (wait) {
2964 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
2965 }
2966
2967 return true;
2968 }
2969
2970 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
2971 double elapsed_time, uint64_t *wait)
2972 {
2973 uint64_t iops_limit = 0;
2974 double ios_limit, ios_base;
2975 double slice_time, wait_time;
2976
2977 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
2978 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2979 } else if (bs->io_limits.iops[is_write]) {
2980 iops_limit = bs->io_limits.iops[is_write];
2981 } else {
2982 if (wait) {
2983 *wait = 0;
2984 }
2985
2986 return false;
2987 }
2988
2989 slice_time = bs->slice_end - bs->slice_start;
2990 slice_time /= (NANOSECONDS_PER_SECOND);
2991 ios_limit = iops_limit * slice_time;
2992 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
2993 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
2994 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
2995 }
2996
2997 if (ios_base + 1 <= ios_limit) {
2998 if (wait) {
2999 *wait = 0;
3000 }
3001
3002 return false;
3003 }
3004
3005 /* Calc approx time to dispatch */
3006 wait_time = (ios_base + 1) / iops_limit;
3007 if (wait_time > elapsed_time) {
3008 wait_time = wait_time - elapsed_time;
3009 } else {
3010 wait_time = 0;
3011 }
3012
3013 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3014 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3015 if (wait) {
3016 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3017 }
3018
3019 return true;
3020 }
3021
3022 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3023 bool is_write, int64_t *wait)
3024 {
3025 int64_t now, max_wait;
3026 uint64_t bps_wait = 0, iops_wait = 0;
3027 double elapsed_time;
3028 int bps_ret, iops_ret;
3029
3030 now = qemu_get_clock_ns(vm_clock);
3031 if ((bs->slice_start < now)
3032 && (bs->slice_end > now)) {
3033 bs->slice_end = now + bs->slice_time;
3034 } else {
3035 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3036 bs->slice_start = now;
3037 bs->slice_end = now + bs->slice_time;
3038
3039 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3040 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3041
3042 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3043 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3044 }
3045
3046 elapsed_time = now - bs->slice_start;
3047 elapsed_time /= (NANOSECONDS_PER_SECOND);
3048
3049 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3050 is_write, elapsed_time, &bps_wait);
3051 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3052 elapsed_time, &iops_wait);
3053 if (bps_ret || iops_ret) {
3054 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3055 if (wait) {
3056 *wait = max_wait;
3057 }
3058
3059 now = qemu_get_clock_ns(vm_clock);
3060 if (bs->slice_end < now + max_wait) {
3061 bs->slice_end = now + max_wait;
3062 }
3063
3064 return true;
3065 }
3066
3067 if (wait) {
3068 *wait = 0;
3069 }
3070
3071 return false;
3072 }
3073
3074 /**************************************************************/
3075 /* async block device emulation */
3076
3077 typedef struct BlockDriverAIOCBSync {
3078 BlockDriverAIOCB common;
3079 QEMUBH *bh;
3080 int ret;
3081 /* vector translation state */
3082 QEMUIOVector *qiov;
3083 uint8_t *bounce;
3084 int is_write;
3085 } BlockDriverAIOCBSync;
3086
3087 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3088 {
3089 BlockDriverAIOCBSync *acb =
3090 container_of(blockacb, BlockDriverAIOCBSync, common);
3091 qemu_bh_delete(acb->bh);
3092 acb->bh = NULL;
3093 qemu_aio_release(acb);
3094 }
3095
3096 static AIOPool bdrv_em_aio_pool = {
3097 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3098 .cancel = bdrv_aio_cancel_em,
3099 };
3100
3101 static void bdrv_aio_bh_cb(void *opaque)
3102 {
3103 BlockDriverAIOCBSync *acb = opaque;
3104
3105 if (!acb->is_write)
3106 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3107 qemu_vfree(acb->bounce);
3108 acb->common.cb(acb->common.opaque, acb->ret);
3109 qemu_bh_delete(acb->bh);
3110 acb->bh = NULL;
3111 qemu_aio_release(acb);
3112 }
3113
3114 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3115 int64_t sector_num,
3116 QEMUIOVector *qiov,
3117 int nb_sectors,
3118 BlockDriverCompletionFunc *cb,
3119 void *opaque,
3120 int is_write)
3121
3122 {
3123 BlockDriverAIOCBSync *acb;
3124
3125 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3126 acb->is_write = is_write;
3127 acb->qiov = qiov;
3128 acb->bounce = qemu_blockalign(bs, qiov->size);
3129 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3130
3131 if (is_write) {
3132 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3133 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3134 } else {
3135 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3136 }
3137
3138 qemu_bh_schedule(acb->bh);
3139
3140 return &acb->common;
3141 }
3142
3143 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3144 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3145 BlockDriverCompletionFunc *cb, void *opaque)
3146 {
3147 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3148 }
3149
3150 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3151 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3152 BlockDriverCompletionFunc *cb, void *opaque)
3153 {
3154 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3155 }
3156
3157
3158 typedef struct BlockDriverAIOCBCoroutine {
3159 BlockDriverAIOCB common;
3160 BlockRequest req;
3161 bool is_write;
3162 QEMUBH* bh;
3163 } BlockDriverAIOCBCoroutine;
3164
3165 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3166 {
3167 qemu_aio_flush();
3168 }
3169
3170 static AIOPool bdrv_em_co_aio_pool = {
3171 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3172 .cancel = bdrv_aio_co_cancel_em,
3173 };
3174
3175 static void bdrv_co_em_bh(void *opaque)
3176 {
3177 BlockDriverAIOCBCoroutine *acb = opaque;
3178
3179 acb->common.cb(acb->common.opaque, acb->req.error);
3180 qemu_bh_delete(acb->bh);
3181 qemu_aio_release(acb);
3182 }
3183
3184 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3185 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3186 {
3187 BlockDriverAIOCBCoroutine *acb = opaque;
3188 BlockDriverState *bs = acb->common.bs;
3189
3190 if (!acb->is_write) {
3191 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3192 acb->req.nb_sectors, acb->req.qiov, 0);
3193 } else {
3194 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3195 acb->req.nb_sectors, acb->req.qiov);
3196 }
3197
3198 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3199 qemu_bh_schedule(acb->bh);
3200 }
3201
3202 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3203 int64_t sector_num,
3204 QEMUIOVector *qiov,
3205 int nb_sectors,
3206 BlockDriverCompletionFunc *cb,
3207 void *opaque,
3208 bool is_write)
3209 {
3210 Coroutine *co;
3211 BlockDriverAIOCBCoroutine *acb;
3212
3213 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3214 acb->req.sector = sector_num;
3215 acb->req.nb_sectors = nb_sectors;
3216 acb->req.qiov = qiov;
3217 acb->is_write = is_write;
3218
3219 co = qemu_coroutine_create(bdrv_co_do_rw);
3220 qemu_coroutine_enter(co, acb);
3221
3222 return &acb->common;
3223 }
3224
3225 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3226 {
3227 BlockDriverAIOCBCoroutine *acb = opaque;
3228 BlockDriverState *bs = acb->common.bs;
3229
3230 acb->req.error = bdrv_co_flush(bs);
3231 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3232 qemu_bh_schedule(acb->bh);
3233 }
3234
3235 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3236 BlockDriverCompletionFunc *cb, void *opaque)
3237 {
3238 trace_bdrv_aio_flush(bs, opaque);
3239
3240 Coroutine *co;
3241 BlockDriverAIOCBCoroutine *acb;
3242
3243 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3244 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3245 qemu_coroutine_enter(co, acb);
3246
3247 return &acb->common;
3248 }
3249
3250 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3251 {
3252 BlockDriverAIOCBCoroutine *acb = opaque;
3253 BlockDriverState *bs = acb->common.bs;
3254
3255 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3256 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3257 qemu_bh_schedule(acb->bh);
3258 }
3259
3260 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3261 int64_t sector_num, int nb_sectors,
3262 BlockDriverCompletionFunc *cb, void *opaque)
3263 {
3264 Coroutine *co;
3265 BlockDriverAIOCBCoroutine *acb;
3266
3267 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3268
3269 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3270 acb->req.sector = sector_num;
3271 acb->req.nb_sectors = nb_sectors;
3272 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3273 qemu_coroutine_enter(co, acb);
3274
3275 return &acb->common;
3276 }
3277
3278 void bdrv_init(void)
3279 {
3280 module_call_init(MODULE_INIT_BLOCK);
3281 }
3282
3283 void bdrv_init_with_whitelist(void)
3284 {
3285 use_bdrv_whitelist = 1;
3286 bdrv_init();
3287 }
3288
3289 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3290 BlockDriverCompletionFunc *cb, void *opaque)
3291 {
3292 BlockDriverAIOCB *acb;
3293
3294 if (pool->free_aiocb) {
3295 acb = pool->free_aiocb;
3296 pool->free_aiocb = acb->next;
3297 } else {
3298 acb = g_malloc0(pool->aiocb_size);
3299 acb->pool = pool;
3300 }
3301 acb->bs = bs;
3302 acb->cb = cb;
3303 acb->opaque = opaque;
3304 return acb;
3305 }
3306
3307 void qemu_aio_release(void *p)
3308 {
3309 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3310 AIOPool *pool = acb->pool;
3311 acb->next = pool->free_aiocb;
3312 pool->free_aiocb = acb;
3313 }
3314
3315 /**************************************************************/
3316 /* Coroutine block device emulation */
3317
3318 typedef struct CoroutineIOCompletion {
3319 Coroutine *coroutine;
3320 int ret;
3321 } CoroutineIOCompletion;
3322
3323 static void bdrv_co_io_em_complete(void *opaque, int ret)
3324 {
3325 CoroutineIOCompletion *co = opaque;
3326
3327 co->ret = ret;
3328 qemu_coroutine_enter(co->coroutine, NULL);
3329 }
3330
3331 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3332 int nb_sectors, QEMUIOVector *iov,
3333 bool is_write)
3334 {
3335 CoroutineIOCompletion co = {
3336 .coroutine = qemu_coroutine_self(),
3337 };
3338 BlockDriverAIOCB *acb;
3339
3340 if (is_write) {
3341 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3342 bdrv_co_io_em_complete, &co);
3343 } else {
3344 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3345 bdrv_co_io_em_complete, &co);
3346 }
3347
3348 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3349 if (!acb) {
3350 return -EIO;
3351 }
3352 qemu_coroutine_yield();
3353
3354 return co.ret;
3355 }
3356
3357 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3358 int64_t sector_num, int nb_sectors,
3359 QEMUIOVector *iov)
3360 {
3361 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3362 }
3363
3364 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3365 int64_t sector_num, int nb_sectors,
3366 QEMUIOVector *iov)
3367 {
3368 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3369 }
3370
3371 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3372 {
3373 RwCo *rwco = opaque;
3374
3375 rwco->ret = bdrv_co_flush(rwco->bs);
3376 }
3377
3378 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3379 {
3380 int ret;
3381
3382 if (!bs->drv) {
3383 return 0;
3384 }
3385
3386 /* Write back cached data to the OS even with cache=unsafe */
3387 if (bs->drv->bdrv_co_flush_to_os) {
3388 ret = bs->drv->bdrv_co_flush_to_os(bs);
3389 if (ret < 0) {
3390 return ret;
3391 }
3392 }
3393
3394 /* But don't actually force it to the disk with cache=unsafe */
3395 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3396 return 0;
3397 }
3398
3399 if (bs->drv->bdrv_co_flush_to_disk) {
3400 return bs->drv->bdrv_co_flush_to_disk(bs);
3401 } else if (bs->drv->bdrv_aio_flush) {
3402 BlockDriverAIOCB *acb;
3403 CoroutineIOCompletion co = {
3404 .coroutine = qemu_coroutine_self(),
3405 };
3406
3407 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3408 if (acb == NULL) {
3409 return -EIO;
3410 } else {
3411 qemu_coroutine_yield();
3412 return co.ret;
3413 }
3414 } else {
3415 /*
3416 * Some block drivers always operate in either writethrough or unsafe
3417 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3418 * know how the server works (because the behaviour is hardcoded or
3419 * depends on server-side configuration), so we can't ensure that
3420 * everything is safe on disk. Returning an error doesn't work because
3421 * that would break guests even if the server operates in writethrough
3422 * mode.
3423 *
3424 * Let's hope the user knows what he's doing.
3425 */
3426 return 0;
3427 }
3428 }
3429
3430 void bdrv_invalidate_cache(BlockDriverState *bs)
3431 {
3432 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3433 bs->drv->bdrv_invalidate_cache(bs);
3434 }
3435 }
3436
3437 void bdrv_invalidate_cache_all(void)
3438 {
3439 BlockDriverState *bs;
3440
3441 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3442 bdrv_invalidate_cache(bs);
3443 }
3444 }
3445
3446 int bdrv_flush(BlockDriverState *bs)
3447 {
3448 Coroutine *co;
3449 RwCo rwco = {
3450 .bs = bs,
3451 .ret = NOT_DONE,
3452 };
3453
3454 if (qemu_in_coroutine()) {
3455 /* Fast-path if already in coroutine context */
3456 bdrv_flush_co_entry(&rwco);
3457 } else {
3458 co = qemu_coroutine_create(bdrv_flush_co_entry);
3459 qemu_coroutine_enter(co, &rwco);
3460 while (rwco.ret == NOT_DONE) {
3461 qemu_aio_wait();
3462 }
3463 }
3464
3465 return rwco.ret;
3466 }
3467
3468 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3469 {
3470 RwCo *rwco = opaque;
3471
3472 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3473 }
3474
3475 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3476 int nb_sectors)
3477 {
3478 if (!bs->drv) {
3479 return -ENOMEDIUM;
3480 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3481 return -EIO;
3482 } else if (bs->read_only) {
3483 return -EROFS;
3484 } else if (bs->drv->bdrv_co_discard) {
3485 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3486 } else if (bs->drv->bdrv_aio_discard) {
3487 BlockDriverAIOCB *acb;
3488 CoroutineIOCompletion co = {
3489 .coroutine = qemu_coroutine_self(),
3490 };
3491
3492 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3493 bdrv_co_io_em_complete, &co);
3494 if (acb == NULL) {
3495 return -EIO;
3496 } else {
3497 qemu_coroutine_yield();
3498 return co.ret;
3499 }
3500 } else {
3501 return 0;
3502 }
3503 }
3504
3505 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3506 {
3507 Coroutine *co;
3508 RwCo rwco = {
3509 .bs = bs,
3510 .sector_num = sector_num,
3511 .nb_sectors = nb_sectors,
3512 .ret = NOT_DONE,
3513 };
3514
3515 if (qemu_in_coroutine()) {
3516 /* Fast-path if already in coroutine context */
3517 bdrv_discard_co_entry(&rwco);
3518 } else {
3519 co = qemu_coroutine_create(bdrv_discard_co_entry);
3520 qemu_coroutine_enter(co, &rwco);
3521 while (rwco.ret == NOT_DONE) {
3522 qemu_aio_wait();
3523 }
3524 }
3525
3526 return rwco.ret;
3527 }
3528
3529 /**************************************************************/
3530 /* removable device support */
3531
3532 /**
3533 * Return TRUE if the media is present
3534 */
3535 int bdrv_is_inserted(BlockDriverState *bs)
3536 {
3537 BlockDriver *drv = bs->drv;
3538
3539 if (!drv)
3540 return 0;
3541 if (!drv->bdrv_is_inserted)
3542 return 1;
3543 return drv->bdrv_is_inserted(bs);
3544 }
3545
3546 /**
3547 * Return whether the media changed since the last call to this
3548 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3549 */
3550 int bdrv_media_changed(BlockDriverState *bs)
3551 {
3552 BlockDriver *drv = bs->drv;
3553
3554 if (drv && drv->bdrv_media_changed) {
3555 return drv->bdrv_media_changed(bs);
3556 }
3557 return -ENOTSUP;
3558 }
3559
3560 /**
3561 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3562 */
3563 void bdrv_eject(BlockDriverState *bs, int eject_flag)
3564 {
3565 BlockDriver *drv = bs->drv;
3566
3567 if (drv && drv->bdrv_eject) {
3568 drv->bdrv_eject(bs, eject_flag);
3569 }
3570 }
3571
3572 /**
3573 * Lock or unlock the media (if it is locked, the user won't be able
3574 * to eject it manually).
3575 */
3576 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3577 {
3578 BlockDriver *drv = bs->drv;
3579
3580 trace_bdrv_lock_medium(bs, locked);
3581
3582 if (drv && drv->bdrv_lock_medium) {
3583 drv->bdrv_lock_medium(bs, locked);
3584 }
3585 }
3586
3587 /* needed for generic scsi interface */
3588
3589 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3590 {
3591 BlockDriver *drv = bs->drv;
3592
3593 if (drv && drv->bdrv_ioctl)
3594 return drv->bdrv_ioctl(bs, req, buf);
3595 return -ENOTSUP;
3596 }
3597
3598 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3599 unsigned long int req, void *buf,
3600 BlockDriverCompletionFunc *cb, void *opaque)
3601 {
3602 BlockDriver *drv = bs->drv;
3603
3604 if (drv && drv->bdrv_aio_ioctl)
3605 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3606 return NULL;
3607 }
3608
3609 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3610 {
3611 bs->buffer_alignment = align;
3612 }
3613
3614 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3615 {
3616 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3617 }
3618
3619 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3620 {
3621 int64_t bitmap_size;
3622
3623 bs->dirty_count = 0;
3624 if (enable) {
3625 if (!bs->dirty_bitmap) {
3626 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3627 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3628 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3629
3630 bs->dirty_bitmap = g_malloc0(bitmap_size);
3631 }
3632 } else {
3633 if (bs->dirty_bitmap) {
3634 g_free(bs->dirty_bitmap);
3635 bs->dirty_bitmap = NULL;
3636 }
3637 }
3638 }
3639
3640 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3641 {
3642 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3643
3644 if (bs->dirty_bitmap &&
3645 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3646 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3647 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3648 } else {
3649 return 0;
3650 }
3651 }
3652
3653 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3654 int nr_sectors)
3655 {
3656 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3657 }
3658
3659 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3660 {
3661 return bs->dirty_count;
3662 }
3663
3664 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3665 {
3666 assert(bs->in_use != in_use);
3667 bs->in_use = in_use;
3668 }
3669
3670 int bdrv_in_use(BlockDriverState *bs)
3671 {
3672 return bs->in_use;
3673 }
3674
3675 void bdrv_iostatus_enable(BlockDriverState *bs)
3676 {
3677 bs->iostatus_enabled = true;
3678 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3679 }
3680
3681 /* The I/O status is only enabled if the drive explicitly
3682 * enables it _and_ the VM is configured to stop on errors */
3683 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3684 {
3685 return (bs->iostatus_enabled &&
3686 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3687 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3688 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3689 }
3690
3691 void bdrv_iostatus_disable(BlockDriverState *bs)
3692 {
3693 bs->iostatus_enabled = false;
3694 }
3695
3696 void bdrv_iostatus_reset(BlockDriverState *bs)
3697 {
3698 if (bdrv_iostatus_is_enabled(bs)) {
3699 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3700 }
3701 }
3702
3703 /* XXX: Today this is set by device models because it makes the implementation
3704 quite simple. However, the block layer knows about the error, so it's
3705 possible to implement this without device models being involved */
3706 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3707 {
3708 if (bdrv_iostatus_is_enabled(bs) &&
3709 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3710 assert(error >= 0);
3711 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
3712 BLOCK_DEVICE_IO_STATUS_FAILED;
3713 }
3714 }
3715
3716 void
3717 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3718 enum BlockAcctType type)
3719 {
3720 assert(type < BDRV_MAX_IOTYPE);
3721
3722 cookie->bytes = bytes;
3723 cookie->start_time_ns = get_clock();
3724 cookie->type = type;
3725 }
3726
3727 void
3728 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3729 {
3730 assert(cookie->type < BDRV_MAX_IOTYPE);
3731
3732 bs->nr_bytes[cookie->type] += cookie->bytes;
3733 bs->nr_ops[cookie->type]++;
3734 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
3735 }
3736
3737 int bdrv_img_create(const char *filename, const char *fmt,
3738 const char *base_filename, const char *base_fmt,
3739 char *options, uint64_t img_size, int flags)
3740 {
3741 QEMUOptionParameter *param = NULL, *create_options = NULL;
3742 QEMUOptionParameter *backing_fmt, *backing_file, *size;
3743 BlockDriverState *bs = NULL;
3744 BlockDriver *drv, *proto_drv;
3745 BlockDriver *backing_drv = NULL;
3746 int ret = 0;
3747
3748 /* Find driver and parse its options */
3749 drv = bdrv_find_format(fmt);
3750 if (!drv) {
3751 error_report("Unknown file format '%s'", fmt);
3752 ret = -EINVAL;
3753 goto out;
3754 }
3755
3756 proto_drv = bdrv_find_protocol(filename);
3757 if (!proto_drv) {
3758 error_report("Unknown protocol '%s'", filename);
3759 ret = -EINVAL;
3760 goto out;
3761 }
3762
3763 create_options = append_option_parameters(create_options,
3764 drv->create_options);
3765 create_options = append_option_parameters(create_options,
3766 proto_drv->create_options);
3767
3768 /* Create parameter list with default values */
3769 param = parse_option_parameters("", create_options, param);
3770
3771 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3772
3773 /* Parse -o options */
3774 if (options) {
3775 param = parse_option_parameters(options, create_options, param);
3776 if (param == NULL) {
3777 error_report("Invalid options for file format '%s'.", fmt);
3778 ret = -EINVAL;
3779 goto out;
3780 }
3781 }
3782
3783 if (base_filename) {
3784 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3785 base_filename)) {
3786 error_report("Backing file not supported for file format '%s'",
3787 fmt);
3788 ret = -EINVAL;
3789 goto out;
3790 }
3791 }
3792
3793 if (base_fmt) {
3794 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
3795 error_report("Backing file format not supported for file "
3796 "format '%s'", fmt);
3797 ret = -EINVAL;
3798 goto out;
3799 }
3800 }
3801
3802 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
3803 if (backing_file && backing_file->value.s) {
3804 if (!strcmp(filename, backing_file->value.s)) {
3805 error_report("Error: Trying to create an image with the "
3806 "same filename as the backing file");
3807 ret = -EINVAL;
3808 goto out;
3809 }
3810 }
3811
3812 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
3813 if (backing_fmt && backing_fmt->value.s) {
3814 backing_drv = bdrv_find_format(backing_fmt->value.s);
3815 if (!backing_drv) {
3816 error_report("Unknown backing file format '%s'",
3817 backing_fmt->value.s);
3818 ret = -EINVAL;
3819 goto out;
3820 }
3821 }
3822
3823 // The size for the image must always be specified, with one exception:
3824 // If we are using a backing file, we can obtain the size from there
3825 size = get_option_parameter(param, BLOCK_OPT_SIZE);
3826 if (size && size->value.n == -1) {
3827 if (backing_file && backing_file->value.s) {
3828 uint64_t size;
3829 char buf[32];
3830
3831 bs = bdrv_new("");
3832
3833 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
3834 if (ret < 0) {
3835 error_report("Could not open '%s'", backing_file->value.s);
3836 goto out;
3837 }
3838 bdrv_get_geometry(bs, &size);
3839 size *= 512;
3840
3841 snprintf(buf, sizeof(buf), "%" PRId64, size);
3842 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
3843 } else {
3844 error_report("Image creation needs a size parameter");
3845 ret = -EINVAL;
3846 goto out;
3847 }
3848 }
3849
3850 printf("Formatting '%s', fmt=%s ", filename, fmt);
3851 print_option_parameters(param);
3852 puts("");
3853
3854 ret = bdrv_create(drv, filename, param);
3855
3856 if (ret < 0) {
3857 if (ret == -ENOTSUP) {
3858 error_report("Formatting or formatting option not supported for "
3859 "file format '%s'", fmt);
3860 } else if (ret == -EFBIG) {
3861 error_report("The image size is too large for file format '%s'",
3862 fmt);
3863 } else {
3864 error_report("%s: error while creating %s: %s", filename, fmt,
3865 strerror(-ret));
3866 }
3867 }
3868
3869 out:
3870 free_option_parameters(create_options);
3871 free_option_parameters(param);
3872
3873 if (bs) {
3874 bdrv_delete(bs);
3875 }
3876
3877 return ret;
3878 }
3879
3880 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
3881 BlockDriverCompletionFunc *cb, void *opaque)
3882 {
3883 BlockJob *job;
3884
3885 if (bs->job || bdrv_in_use(bs)) {
3886 return NULL;
3887 }
3888 bdrv_set_in_use(bs, 1);
3889
3890 job = g_malloc0(job_type->instance_size);
3891 job->job_type = job_type;
3892 job->bs = bs;
3893 job->cb = cb;
3894 job->opaque = opaque;
3895 bs->job = job;
3896 return job;
3897 }
3898
3899 void block_job_complete(BlockJob *job, int ret)
3900 {
3901 BlockDriverState *bs = job->bs;
3902
3903 assert(bs->job == job);
3904 job->cb(job->opaque, ret);
3905 bs->job = NULL;
3906 g_free(job);
3907 bdrv_set_in_use(bs, 0);
3908 }
3909
3910 int block_job_set_speed(BlockJob *job, int64_t value)
3911 {
3912 if (!job->job_type->set_speed) {
3913 return -ENOTSUP;
3914 }
3915 return job->job_type->set_speed(job, value);
3916 }
3917
3918 void block_job_cancel(BlockJob *job)
3919 {
3920 job->cancelled = true;
3921 }
3922
3923 bool block_job_is_cancelled(BlockJob *job)
3924 {
3925 return job->cancelled;
3926 }
Random patches