root / block.c @ dc1c13d9
History | View | Annotate | Download (118.8 kB)
1 |
/*
|
---|---|
2 |
* QEMU System Emulator block driver
|
3 |
*
|
4 |
* Copyright (c) 2003 Fabrice Bellard
|
5 |
*
|
6 |
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
7 |
* of this software and associated documentation files (the "Software"), to deal
|
8 |
* in the Software without restriction, including without limitation the rights
|
9 |
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
10 |
* copies of the Software, and to permit persons to whom the Software is
|
11 |
* furnished to do so, subject to the following conditions:
|
12 |
*
|
13 |
* The above copyright notice and this permission notice shall be included in
|
14 |
* all copies or substantial portions of the Software.
|
15 |
*
|
16 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
17 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
18 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
19 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
20 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
21 |
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
22 |
* THE SOFTWARE.
|
23 |
*/
|
24 |
#include "config-host.h" |
25 |
#include "qemu-common.h" |
26 |
#include "trace.h" |
27 |
#include "monitor.h" |
28 |
#include "block_int.h" |
29 |
#include "module.h" |
30 |
#include "qjson.h" |
31 |
#include "qemu-coroutine.h" |
32 |
#include "qmp-commands.h" |
33 |
#include "qemu-timer.h" |
34 |
|
35 |
#ifdef CONFIG_BSD
|
36 |
#include <sys/types.h> |
37 |
#include <sys/stat.h> |
38 |
#include <sys/ioctl.h> |
39 |
#include <sys/queue.h> |
40 |
#ifndef __DragonFly__
|
41 |
#include <sys/disk.h> |
42 |
#endif
|
43 |
#endif
|
44 |
|
45 |
#ifdef _WIN32
|
46 |
#include <windows.h> |
47 |
#endif
|
48 |
|
49 |
#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ |
50 |
|
51 |
typedef enum { |
52 |
BDRV_REQ_COPY_ON_READ = 0x1,
|
53 |
BDRV_REQ_ZERO_WRITE = 0x2,
|
54 |
} BdrvRequestFlags; |
55 |
|
56 |
static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load); |
57 |
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
|
58 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
59 |
BlockDriverCompletionFunc *cb, void *opaque);
|
60 |
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
|
61 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
62 |
BlockDriverCompletionFunc *cb, void *opaque);
|
63 |
static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, |
64 |
int64_t sector_num, int nb_sectors,
|
65 |
QEMUIOVector *iov); |
66 |
static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, |
67 |
int64_t sector_num, int nb_sectors,
|
68 |
QEMUIOVector *iov); |
69 |
static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, |
70 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
71 |
BdrvRequestFlags flags); |
72 |
static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, |
73 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
74 |
BdrvRequestFlags flags); |
75 |
static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
|
76 |
int64_t sector_num, |
77 |
QEMUIOVector *qiov, |
78 |
int nb_sectors,
|
79 |
BlockDriverCompletionFunc *cb, |
80 |
void *opaque,
|
81 |
bool is_write);
|
82 |
static void coroutine_fn bdrv_co_do_rw(void *opaque); |
83 |
static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, |
84 |
int64_t sector_num, int nb_sectors);
|
85 |
|
86 |
static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors, |
87 |
bool is_write, double elapsed_time, uint64_t *wait); |
88 |
static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write, |
89 |
double elapsed_time, uint64_t *wait);
|
90 |
static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors, |
91 |
bool is_write, int64_t *wait);
|
92 |
|
93 |
static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
|
94 |
QTAILQ_HEAD_INITIALIZER(bdrv_states); |
95 |
|
96 |
static QLIST_HEAD(, BlockDriver) bdrv_drivers =
|
97 |
QLIST_HEAD_INITIALIZER(bdrv_drivers); |
98 |
|
99 |
/* The device to use for VM snapshots */
|
100 |
static BlockDriverState *bs_snapshots;
|
101 |
|
102 |
/* If non-zero, use only whitelisted block drivers */
|
103 |
static int use_bdrv_whitelist; |
104 |
|
105 |
#ifdef _WIN32
|
106 |
static int is_windows_drive_prefix(const char *filename) |
107 |
{ |
108 |
return (((filename[0] >= 'a' && filename[0] <= 'z') || |
109 |
(filename[0] >= 'A' && filename[0] <= 'Z')) && |
110 |
filename[1] == ':'); |
111 |
} |
112 |
|
113 |
int is_windows_drive(const char *filename) |
114 |
{ |
115 |
if (is_windows_drive_prefix(filename) &&
|
116 |
filename[2] == '\0') |
117 |
return 1; |
118 |
if (strstart(filename, "\\\\.\\", NULL) || |
119 |
strstart(filename, "//./", NULL)) |
120 |
return 1; |
121 |
return 0; |
122 |
} |
123 |
#endif
|
124 |
|
125 |
/* throttling disk I/O limits */
|
126 |
void bdrv_io_limits_disable(BlockDriverState *bs)
|
127 |
{ |
128 |
bs->io_limits_enabled = false;
|
129 |
|
130 |
while (qemu_co_queue_next(&bs->throttled_reqs));
|
131 |
|
132 |
if (bs->block_timer) {
|
133 |
qemu_del_timer(bs->block_timer); |
134 |
qemu_free_timer(bs->block_timer); |
135 |
bs->block_timer = NULL;
|
136 |
} |
137 |
|
138 |
bs->slice_start = 0;
|
139 |
bs->slice_end = 0;
|
140 |
bs->slice_time = 0;
|
141 |
memset(&bs->io_base, 0, sizeof(bs->io_base)); |
142 |
} |
143 |
|
144 |
static void bdrv_block_timer(void *opaque) |
145 |
{ |
146 |
BlockDriverState *bs = opaque; |
147 |
|
148 |
qemu_co_queue_next(&bs->throttled_reqs); |
149 |
} |
150 |
|
151 |
void bdrv_io_limits_enable(BlockDriverState *bs)
|
152 |
{ |
153 |
qemu_co_queue_init(&bs->throttled_reqs); |
154 |
bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs); |
155 |
bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
|
156 |
bs->slice_start = qemu_get_clock_ns(vm_clock); |
157 |
bs->slice_end = bs->slice_start + bs->slice_time; |
158 |
memset(&bs->io_base, 0, sizeof(bs->io_base)); |
159 |
bs->io_limits_enabled = true;
|
160 |
} |
161 |
|
162 |
bool bdrv_io_limits_enabled(BlockDriverState *bs)
|
163 |
{ |
164 |
BlockIOLimit *io_limits = &bs->io_limits; |
165 |
return io_limits->bps[BLOCK_IO_LIMIT_READ]
|
166 |
|| io_limits->bps[BLOCK_IO_LIMIT_WRITE] |
167 |
|| io_limits->bps[BLOCK_IO_LIMIT_TOTAL] |
168 |
|| io_limits->iops[BLOCK_IO_LIMIT_READ] |
169 |
|| io_limits->iops[BLOCK_IO_LIMIT_WRITE] |
170 |
|| io_limits->iops[BLOCK_IO_LIMIT_TOTAL]; |
171 |
} |
172 |
|
173 |
static void bdrv_io_limits_intercept(BlockDriverState *bs, |
174 |
bool is_write, int nb_sectors) |
175 |
{ |
176 |
int64_t wait_time = -1;
|
177 |
|
178 |
if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
|
179 |
qemu_co_queue_wait(&bs->throttled_reqs); |
180 |
} |
181 |
|
182 |
/* In fact, we hope to keep each request's timing, in FIFO mode. The next
|
183 |
* throttled requests will not be dequeued until the current request is
|
184 |
* allowed to be serviced. So if the current request still exceeds the
|
185 |
* limits, it will be inserted to the head. All requests followed it will
|
186 |
* be still in throttled_reqs queue.
|
187 |
*/
|
188 |
|
189 |
while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
|
190 |
qemu_mod_timer(bs->block_timer, |
191 |
wait_time + qemu_get_clock_ns(vm_clock)); |
192 |
qemu_co_queue_wait_insert_head(&bs->throttled_reqs); |
193 |
} |
194 |
|
195 |
qemu_co_queue_next(&bs->throttled_reqs); |
196 |
} |
197 |
|
198 |
/* check if the path starts with "<protocol>:" */
|
199 |
static int path_has_protocol(const char *path) |
200 |
{ |
201 |
const char *p; |
202 |
|
203 |
#ifdef _WIN32
|
204 |
if (is_windows_drive(path) ||
|
205 |
is_windows_drive_prefix(path)) { |
206 |
return 0; |
207 |
} |
208 |
p = path + strcspn(path, ":/\\");
|
209 |
#else
|
210 |
p = path + strcspn(path, ":/");
|
211 |
#endif
|
212 |
|
213 |
return *p == ':'; |
214 |
} |
215 |
|
216 |
int path_is_absolute(const char *path) |
217 |
{ |
218 |
#ifdef _WIN32
|
219 |
/* specific case for names like: "\\.\d:" */
|
220 |
if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
|
221 |
return 1; |
222 |
} |
223 |
return (*path == '/' || *path == '\\'); |
224 |
#else
|
225 |
return (*path == '/'); |
226 |
#endif
|
227 |
} |
228 |
|
229 |
/* if filename is absolute, just copy it to dest. Otherwise, build a
|
230 |
path to it by considering it is relative to base_path. URL are
|
231 |
supported. */
|
232 |
void path_combine(char *dest, int dest_size, |
233 |
const char *base_path, |
234 |
const char *filename) |
235 |
{ |
236 |
const char *p, *p1; |
237 |
int len;
|
238 |
|
239 |
if (dest_size <= 0) |
240 |
return;
|
241 |
if (path_is_absolute(filename)) {
|
242 |
pstrcpy(dest, dest_size, filename); |
243 |
} else {
|
244 |
p = strchr(base_path, ':');
|
245 |
if (p)
|
246 |
p++; |
247 |
else
|
248 |
p = base_path; |
249 |
p1 = strrchr(base_path, '/');
|
250 |
#ifdef _WIN32
|
251 |
{ |
252 |
const char *p2; |
253 |
p2 = strrchr(base_path, '\\');
|
254 |
if (!p1 || p2 > p1)
|
255 |
p1 = p2; |
256 |
} |
257 |
#endif
|
258 |
if (p1)
|
259 |
p1++; |
260 |
else
|
261 |
p1 = base_path; |
262 |
if (p1 > p)
|
263 |
p = p1; |
264 |
len = p - base_path; |
265 |
if (len > dest_size - 1) |
266 |
len = dest_size - 1;
|
267 |
memcpy(dest, base_path, len); |
268 |
dest[len] = '\0';
|
269 |
pstrcat(dest, dest_size, filename); |
270 |
} |
271 |
} |
272 |
|
273 |
void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz) |
274 |
{ |
275 |
if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) { |
276 |
pstrcpy(dest, sz, bs->backing_file); |
277 |
} else {
|
278 |
path_combine(dest, sz, bs->filename, bs->backing_file); |
279 |
} |
280 |
} |
281 |
|
282 |
void bdrv_register(BlockDriver *bdrv)
|
283 |
{ |
284 |
/* Block drivers without coroutine functions need emulation */
|
285 |
if (!bdrv->bdrv_co_readv) {
|
286 |
bdrv->bdrv_co_readv = bdrv_co_readv_em; |
287 |
bdrv->bdrv_co_writev = bdrv_co_writev_em; |
288 |
|
289 |
/* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
|
290 |
* the block driver lacks aio we need to emulate that too.
|
291 |
*/
|
292 |
if (!bdrv->bdrv_aio_readv) {
|
293 |
/* add AIO emulation layer */
|
294 |
bdrv->bdrv_aio_readv = bdrv_aio_readv_em; |
295 |
bdrv->bdrv_aio_writev = bdrv_aio_writev_em; |
296 |
} |
297 |
} |
298 |
|
299 |
QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); |
300 |
} |
301 |
|
302 |
/* create a new block device (by default it is empty) */
|
303 |
BlockDriverState *bdrv_new(const char *device_name) |
304 |
{ |
305 |
BlockDriverState *bs; |
306 |
|
307 |
bs = g_malloc0(sizeof(BlockDriverState));
|
308 |
pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
|
309 |
if (device_name[0] != '\0') { |
310 |
QTAILQ_INSERT_TAIL(&bdrv_states, bs, list); |
311 |
} |
312 |
bdrv_iostatus_disable(bs); |
313 |
return bs;
|
314 |
} |
315 |
|
316 |
BlockDriver *bdrv_find_format(const char *format_name) |
317 |
{ |
318 |
BlockDriver *drv1; |
319 |
QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
320 |
if (!strcmp(drv1->format_name, format_name)) {
|
321 |
return drv1;
|
322 |
} |
323 |
} |
324 |
return NULL; |
325 |
} |
326 |
|
327 |
static int bdrv_is_whitelisted(BlockDriver *drv) |
328 |
{ |
329 |
static const char *whitelist[] = { |
330 |
CONFIG_BDRV_WHITELIST |
331 |
}; |
332 |
const char **p; |
333 |
|
334 |
if (!whitelist[0]) |
335 |
return 1; /* no whitelist, anything goes */ |
336 |
|
337 |
for (p = whitelist; *p; p++) {
|
338 |
if (!strcmp(drv->format_name, *p)) {
|
339 |
return 1; |
340 |
} |
341 |
} |
342 |
return 0; |
343 |
} |
344 |
|
345 |
BlockDriver *bdrv_find_whitelisted_format(const char *format_name) |
346 |
{ |
347 |
BlockDriver *drv = bdrv_find_format(format_name); |
348 |
return drv && bdrv_is_whitelisted(drv) ? drv : NULL; |
349 |
} |
350 |
|
351 |
typedef struct CreateCo { |
352 |
BlockDriver *drv; |
353 |
char *filename;
|
354 |
QEMUOptionParameter *options; |
355 |
int ret;
|
356 |
} CreateCo; |
357 |
|
358 |
static void coroutine_fn bdrv_create_co_entry(void *opaque) |
359 |
{ |
360 |
CreateCo *cco = opaque; |
361 |
assert(cco->drv); |
362 |
|
363 |
cco->ret = cco->drv->bdrv_create(cco->filename, cco->options); |
364 |
} |
365 |
|
366 |
int bdrv_create(BlockDriver *drv, const char* filename, |
367 |
QEMUOptionParameter *options) |
368 |
{ |
369 |
int ret;
|
370 |
|
371 |
Coroutine *co; |
372 |
CreateCo cco = { |
373 |
.drv = drv, |
374 |
.filename = g_strdup(filename), |
375 |
.options = options, |
376 |
.ret = NOT_DONE, |
377 |
}; |
378 |
|
379 |
if (!drv->bdrv_create) {
|
380 |
return -ENOTSUP;
|
381 |
} |
382 |
|
383 |
if (qemu_in_coroutine()) {
|
384 |
/* Fast-path if already in coroutine context */
|
385 |
bdrv_create_co_entry(&cco); |
386 |
} else {
|
387 |
co = qemu_coroutine_create(bdrv_create_co_entry); |
388 |
qemu_coroutine_enter(co, &cco); |
389 |
while (cco.ret == NOT_DONE) {
|
390 |
qemu_aio_wait(); |
391 |
} |
392 |
} |
393 |
|
394 |
ret = cco.ret; |
395 |
g_free(cco.filename); |
396 |
|
397 |
return ret;
|
398 |
} |
399 |
|
400 |
int bdrv_create_file(const char* filename, QEMUOptionParameter *options) |
401 |
{ |
402 |
BlockDriver *drv; |
403 |
|
404 |
drv = bdrv_find_protocol(filename); |
405 |
if (drv == NULL) { |
406 |
return -ENOENT;
|
407 |
} |
408 |
|
409 |
return bdrv_create(drv, filename, options);
|
410 |
} |
411 |
|
412 |
/*
|
413 |
* Create a uniquely-named empty temporary file.
|
414 |
* Return 0 upon success, otherwise a negative errno value.
|
415 |
*/
|
416 |
int get_tmp_filename(char *filename, int size) |
417 |
{ |
418 |
#ifdef _WIN32
|
419 |
char temp_dir[MAX_PATH];
|
420 |
/* GetTempFileName requires that its output buffer (4th param)
|
421 |
have length MAX_PATH or greater. */
|
422 |
assert(size >= MAX_PATH); |
423 |
return (GetTempPath(MAX_PATH, temp_dir)
|
424 |
&& GetTempFileName(temp_dir, "qem", 0, filename) |
425 |
? 0 : -GetLastError());
|
426 |
#else
|
427 |
int fd;
|
428 |
const char *tmpdir; |
429 |
tmpdir = getenv("TMPDIR");
|
430 |
if (!tmpdir)
|
431 |
tmpdir = "/tmp";
|
432 |
if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { |
433 |
return -EOVERFLOW;
|
434 |
} |
435 |
fd = mkstemp(filename); |
436 |
if (fd < 0) { |
437 |
return -errno;
|
438 |
} |
439 |
if (close(fd) != 0) { |
440 |
unlink(filename); |
441 |
return -errno;
|
442 |
} |
443 |
return 0; |
444 |
#endif
|
445 |
} |
446 |
|
447 |
/*
|
448 |
* Detect host devices. By convention, /dev/cdrom[N] is always
|
449 |
* recognized as a host CDROM.
|
450 |
*/
|
451 |
static BlockDriver *find_hdev_driver(const char *filename) |
452 |
{ |
453 |
int score_max = 0, score; |
454 |
BlockDriver *drv = NULL, *d;
|
455 |
|
456 |
QLIST_FOREACH(d, &bdrv_drivers, list) { |
457 |
if (d->bdrv_probe_device) {
|
458 |
score = d->bdrv_probe_device(filename); |
459 |
if (score > score_max) {
|
460 |
score_max = score; |
461 |
drv = d; |
462 |
} |
463 |
} |
464 |
} |
465 |
|
466 |
return drv;
|
467 |
} |
468 |
|
469 |
BlockDriver *bdrv_find_protocol(const char *filename) |
470 |
{ |
471 |
BlockDriver *drv1; |
472 |
char protocol[128]; |
473 |
int len;
|
474 |
const char *p; |
475 |
|
476 |
/* TODO Drivers without bdrv_file_open must be specified explicitly */
|
477 |
|
478 |
/*
|
479 |
* XXX(hch): we really should not let host device detection
|
480 |
* override an explicit protocol specification, but moving this
|
481 |
* later breaks access to device names with colons in them.
|
482 |
* Thanks to the brain-dead persistent naming schemes on udev-
|
483 |
* based Linux systems those actually are quite common.
|
484 |
*/
|
485 |
drv1 = find_hdev_driver(filename); |
486 |
if (drv1) {
|
487 |
return drv1;
|
488 |
} |
489 |
|
490 |
if (!path_has_protocol(filename)) {
|
491 |
return bdrv_find_format("file"); |
492 |
} |
493 |
p = strchr(filename, ':');
|
494 |
assert(p != NULL);
|
495 |
len = p - filename; |
496 |
if (len > sizeof(protocol) - 1) |
497 |
len = sizeof(protocol) - 1; |
498 |
memcpy(protocol, filename, len); |
499 |
protocol[len] = '\0';
|
500 |
QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
501 |
if (drv1->protocol_name &&
|
502 |
!strcmp(drv1->protocol_name, protocol)) { |
503 |
return drv1;
|
504 |
} |
505 |
} |
506 |
return NULL; |
507 |
} |
508 |
|
509 |
static int find_image_format(const char *filename, BlockDriver **pdrv) |
510 |
{ |
511 |
int ret, score, score_max;
|
512 |
BlockDriver *drv1, *drv; |
513 |
uint8_t buf[2048];
|
514 |
BlockDriverState *bs; |
515 |
|
516 |
ret = bdrv_file_open(&bs, filename, 0);
|
517 |
if (ret < 0) { |
518 |
*pdrv = NULL;
|
519 |
return ret;
|
520 |
} |
521 |
|
522 |
/* Return the raw BlockDriver * to scsi-generic devices or empty drives */
|
523 |
if (bs->sg || !bdrv_is_inserted(bs)) {
|
524 |
bdrv_delete(bs); |
525 |
drv = bdrv_find_format("raw");
|
526 |
if (!drv) {
|
527 |
ret = -ENOENT; |
528 |
} |
529 |
*pdrv = drv; |
530 |
return ret;
|
531 |
} |
532 |
|
533 |
ret = bdrv_pread(bs, 0, buf, sizeof(buf)); |
534 |
bdrv_delete(bs); |
535 |
if (ret < 0) { |
536 |
*pdrv = NULL;
|
537 |
return ret;
|
538 |
} |
539 |
|
540 |
score_max = 0;
|
541 |
drv = NULL;
|
542 |
QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
543 |
if (drv1->bdrv_probe) {
|
544 |
score = drv1->bdrv_probe(buf, ret, filename); |
545 |
if (score > score_max) {
|
546 |
score_max = score; |
547 |
drv = drv1; |
548 |
} |
549 |
} |
550 |
} |
551 |
if (!drv) {
|
552 |
ret = -ENOENT; |
553 |
} |
554 |
*pdrv = drv; |
555 |
return ret;
|
556 |
} |
557 |
|
558 |
/**
|
559 |
* Set the current 'total_sectors' value
|
560 |
*/
|
561 |
static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) |
562 |
{ |
563 |
BlockDriver *drv = bs->drv; |
564 |
|
565 |
/* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
|
566 |
if (bs->sg)
|
567 |
return 0; |
568 |
|
569 |
/* query actual device if possible, otherwise just trust the hint */
|
570 |
if (drv->bdrv_getlength) {
|
571 |
int64_t length = drv->bdrv_getlength(bs); |
572 |
if (length < 0) { |
573 |
return length;
|
574 |
} |
575 |
hint = length >> BDRV_SECTOR_BITS; |
576 |
} |
577 |
|
578 |
bs->total_sectors = hint; |
579 |
return 0; |
580 |
} |
581 |
|
582 |
/**
|
583 |
* Set open flags for a given cache mode
|
584 |
*
|
585 |
* Return 0 on success, -1 if the cache mode was invalid.
|
586 |
*/
|
587 |
int bdrv_parse_cache_flags(const char *mode, int *flags) |
588 |
{ |
589 |
*flags &= ~BDRV_O_CACHE_MASK; |
590 |
|
591 |
if (!strcmp(mode, "off") || !strcmp(mode, "none")) { |
592 |
*flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB; |
593 |
} else if (!strcmp(mode, "directsync")) { |
594 |
*flags |= BDRV_O_NOCACHE; |
595 |
} else if (!strcmp(mode, "writeback")) { |
596 |
*flags |= BDRV_O_CACHE_WB; |
597 |
} else if (!strcmp(mode, "unsafe")) { |
598 |
*flags |= BDRV_O_CACHE_WB; |
599 |
*flags |= BDRV_O_NO_FLUSH; |
600 |
} else if (!strcmp(mode, "writethrough")) { |
601 |
/* this is the default */
|
602 |
} else {
|
603 |
return -1; |
604 |
} |
605 |
|
606 |
return 0; |
607 |
} |
608 |
|
609 |
/**
|
610 |
* The copy-on-read flag is actually a reference count so multiple users may
|
611 |
* use the feature without worrying about clobbering its previous state.
|
612 |
* Copy-on-read stays enabled until all users have called to disable it.
|
613 |
*/
|
614 |
void bdrv_enable_copy_on_read(BlockDriverState *bs)
|
615 |
{ |
616 |
bs->copy_on_read++; |
617 |
} |
618 |
|
619 |
void bdrv_disable_copy_on_read(BlockDriverState *bs)
|
620 |
{ |
621 |
assert(bs->copy_on_read > 0);
|
622 |
bs->copy_on_read--; |
623 |
} |
624 |
|
625 |
/*
|
626 |
* Common part for opening disk images and files
|
627 |
*/
|
628 |
static int bdrv_open_common(BlockDriverState *bs, const char *filename, |
629 |
int flags, BlockDriver *drv)
|
630 |
{ |
631 |
int ret, open_flags;
|
632 |
|
633 |
assert(drv != NULL);
|
634 |
assert(bs->file == NULL);
|
635 |
|
636 |
trace_bdrv_open_common(bs, filename, flags, drv->format_name); |
637 |
|
638 |
bs->open_flags = flags; |
639 |
bs->buffer_alignment = 512;
|
640 |
|
641 |
assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ |
642 |
if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
|
643 |
bdrv_enable_copy_on_read(bs); |
644 |
} |
645 |
|
646 |
pstrcpy(bs->filename, sizeof(bs->filename), filename);
|
647 |
|
648 |
if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
|
649 |
return -ENOTSUP;
|
650 |
} |
651 |
|
652 |
bs->drv = drv; |
653 |
bs->opaque = g_malloc0(drv->instance_size); |
654 |
|
655 |
bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB); |
656 |
open_flags = flags | BDRV_O_CACHE_WB; |
657 |
|
658 |
/*
|
659 |
* Clear flags that are internal to the block layer before opening the
|
660 |
* image.
|
661 |
*/
|
662 |
open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
663 |
|
664 |
/*
|
665 |
* Snapshots should be writable.
|
666 |
*/
|
667 |
if (bs->is_temporary) {
|
668 |
open_flags |= BDRV_O_RDWR; |
669 |
} |
670 |
|
671 |
bs->read_only = !(open_flags & BDRV_O_RDWR); |
672 |
|
673 |
/* Open the image, either directly or using a protocol */
|
674 |
if (drv->bdrv_file_open) {
|
675 |
ret = drv->bdrv_file_open(bs, filename, open_flags); |
676 |
} else {
|
677 |
ret = bdrv_file_open(&bs->file, filename, open_flags); |
678 |
if (ret >= 0) { |
679 |
ret = drv->bdrv_open(bs, open_flags); |
680 |
} |
681 |
} |
682 |
|
683 |
if (ret < 0) { |
684 |
goto free_and_fail;
|
685 |
} |
686 |
|
687 |
ret = refresh_total_sectors(bs, bs->total_sectors); |
688 |
if (ret < 0) { |
689 |
goto free_and_fail;
|
690 |
} |
691 |
|
692 |
#ifndef _WIN32
|
693 |
if (bs->is_temporary) {
|
694 |
unlink(filename); |
695 |
} |
696 |
#endif
|
697 |
return 0; |
698 |
|
699 |
free_and_fail:
|
700 |
if (bs->file) {
|
701 |
bdrv_delete(bs->file); |
702 |
bs->file = NULL;
|
703 |
} |
704 |
g_free(bs->opaque); |
705 |
bs->opaque = NULL;
|
706 |
bs->drv = NULL;
|
707 |
return ret;
|
708 |
} |
709 |
|
710 |
/*
|
711 |
* Opens a file using a protocol (file, host_device, nbd, ...)
|
712 |
*/
|
713 |
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags) |
714 |
{ |
715 |
BlockDriverState *bs; |
716 |
BlockDriver *drv; |
717 |
int ret;
|
718 |
|
719 |
drv = bdrv_find_protocol(filename); |
720 |
if (!drv) {
|
721 |
return -ENOENT;
|
722 |
} |
723 |
|
724 |
bs = bdrv_new("");
|
725 |
ret = bdrv_open_common(bs, filename, flags, drv); |
726 |
if (ret < 0) { |
727 |
bdrv_delete(bs); |
728 |
return ret;
|
729 |
} |
730 |
bs->growable = 1;
|
731 |
*pbs = bs; |
732 |
return 0; |
733 |
} |
734 |
|
735 |
/*
|
736 |
* Opens a disk image (raw, qcow2, vmdk, ...)
|
737 |
*/
|
738 |
int bdrv_open(BlockDriverState *bs, const char *filename, int flags, |
739 |
BlockDriver *drv) |
740 |
{ |
741 |
int ret;
|
742 |
char tmp_filename[PATH_MAX];
|
743 |
|
744 |
if (flags & BDRV_O_SNAPSHOT) {
|
745 |
BlockDriverState *bs1; |
746 |
int64_t total_size; |
747 |
int is_protocol = 0; |
748 |
BlockDriver *bdrv_qcow2; |
749 |
QEMUOptionParameter *options; |
750 |
char backing_filename[PATH_MAX];
|
751 |
|
752 |
/* if snapshot, we create a temporary backing file and open it
|
753 |
instead of opening 'filename' directly */
|
754 |
|
755 |
/* if there is a backing file, use it */
|
756 |
bs1 = bdrv_new("");
|
757 |
ret = bdrv_open(bs1, filename, 0, drv);
|
758 |
if (ret < 0) { |
759 |
bdrv_delete(bs1); |
760 |
return ret;
|
761 |
} |
762 |
total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK; |
763 |
|
764 |
if (bs1->drv && bs1->drv->protocol_name)
|
765 |
is_protocol = 1;
|
766 |
|
767 |
bdrv_delete(bs1); |
768 |
|
769 |
ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
|
770 |
if (ret < 0) { |
771 |
return ret;
|
772 |
} |
773 |
|
774 |
/* Real path is meaningless for protocols */
|
775 |
if (is_protocol)
|
776 |
snprintf(backing_filename, sizeof(backing_filename),
|
777 |
"%s", filename);
|
778 |
else if (!realpath(filename, backing_filename)) |
779 |
return -errno;
|
780 |
|
781 |
bdrv_qcow2 = bdrv_find_format("qcow2");
|
782 |
options = parse_option_parameters("", bdrv_qcow2->create_options, NULL); |
783 |
|
784 |
set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size); |
785 |
set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename); |
786 |
if (drv) {
|
787 |
set_option_parameter(options, BLOCK_OPT_BACKING_FMT, |
788 |
drv->format_name); |
789 |
} |
790 |
|
791 |
ret = bdrv_create(bdrv_qcow2, tmp_filename, options); |
792 |
free_option_parameters(options); |
793 |
if (ret < 0) { |
794 |
return ret;
|
795 |
} |
796 |
|
797 |
filename = tmp_filename; |
798 |
drv = bdrv_qcow2; |
799 |
bs->is_temporary = 1;
|
800 |
} |
801 |
|
802 |
/* Find the right image format driver */
|
803 |
if (!drv) {
|
804 |
ret = find_image_format(filename, &drv); |
805 |
} |
806 |
|
807 |
if (!drv) {
|
808 |
goto unlink_and_fail;
|
809 |
} |
810 |
|
811 |
if (flags & BDRV_O_RDWR) {
|
812 |
flags |= BDRV_O_ALLOW_RDWR; |
813 |
} |
814 |
|
815 |
/* Open the image */
|
816 |
ret = bdrv_open_common(bs, filename, flags, drv); |
817 |
if (ret < 0) { |
818 |
goto unlink_and_fail;
|
819 |
} |
820 |
|
821 |
/* If there is a backing file, use it */
|
822 |
if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') { |
823 |
char backing_filename[PATH_MAX];
|
824 |
int back_flags;
|
825 |
BlockDriver *back_drv = NULL;
|
826 |
|
827 |
bs->backing_hd = bdrv_new("");
|
828 |
bdrv_get_full_backing_filename(bs, backing_filename, |
829 |
sizeof(backing_filename));
|
830 |
|
831 |
if (bs->backing_format[0] != '\0') { |
832 |
back_drv = bdrv_find_format(bs->backing_format); |
833 |
} |
834 |
|
835 |
/* backing files always opened read-only */
|
836 |
back_flags = |
837 |
flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
838 |
|
839 |
ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv); |
840 |
if (ret < 0) { |
841 |
bdrv_close(bs); |
842 |
return ret;
|
843 |
} |
844 |
} |
845 |
|
846 |
if (!bdrv_key_required(bs)) {
|
847 |
bdrv_dev_change_media_cb(bs, true);
|
848 |
} |
849 |
|
850 |
/* throttling disk I/O limits */
|
851 |
if (bs->io_limits_enabled) {
|
852 |
bdrv_io_limits_enable(bs); |
853 |
} |
854 |
|
855 |
return 0; |
856 |
|
857 |
unlink_and_fail:
|
858 |
if (bs->is_temporary) {
|
859 |
unlink(filename); |
860 |
} |
861 |
return ret;
|
862 |
} |
863 |
|
864 |
typedef struct BlockReopenQueueEntry { |
865 |
bool prepared;
|
866 |
BDRVReopenState state; |
867 |
QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; |
868 |
} BlockReopenQueueEntry; |
869 |
|
870 |
/*
|
871 |
* Adds a BlockDriverState to a simple queue for an atomic, transactional
|
872 |
* reopen of multiple devices.
|
873 |
*
|
874 |
* bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
|
875 |
* already performed, or alternatively may be NULL a new BlockReopenQueue will
|
876 |
* be created and initialized. This newly created BlockReopenQueue should be
|
877 |
* passed back in for subsequent calls that are intended to be of the same
|
878 |
* atomic 'set'.
|
879 |
*
|
880 |
* bs is the BlockDriverState to add to the reopen queue.
|
881 |
*
|
882 |
* flags contains the open flags for the associated bs
|
883 |
*
|
884 |
* returns a pointer to bs_queue, which is either the newly allocated
|
885 |
* bs_queue, or the existing bs_queue being used.
|
886 |
*
|
887 |
*/
|
888 |
BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
889 |
BlockDriverState *bs, int flags)
|
890 |
{ |
891 |
assert(bs != NULL);
|
892 |
|
893 |
BlockReopenQueueEntry *bs_entry; |
894 |
if (bs_queue == NULL) { |
895 |
bs_queue = g_new0(BlockReopenQueue, 1);
|
896 |
QSIMPLEQ_INIT(bs_queue); |
897 |
} |
898 |
|
899 |
if (bs->file) {
|
900 |
bdrv_reopen_queue(bs_queue, bs->file, flags); |
901 |
} |
902 |
|
903 |
bs_entry = g_new0(BlockReopenQueueEntry, 1);
|
904 |
QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
905 |
|
906 |
bs_entry->state.bs = bs; |
907 |
bs_entry->state.flags = flags; |
908 |
|
909 |
return bs_queue;
|
910 |
} |
911 |
|
912 |
/*
|
913 |
* Reopen multiple BlockDriverStates atomically & transactionally.
|
914 |
*
|
915 |
* The queue passed in (bs_queue) must have been built up previous
|
916 |
* via bdrv_reopen_queue().
|
917 |
*
|
918 |
* Reopens all BDS specified in the queue, with the appropriate
|
919 |
* flags. All devices are prepared for reopen, and failure of any
|
920 |
* device will cause all device changes to be abandonded, and intermediate
|
921 |
* data cleaned up.
|
922 |
*
|
923 |
* If all devices prepare successfully, then the changes are committed
|
924 |
* to all devices.
|
925 |
*
|
926 |
*/
|
927 |
int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
|
928 |
{ |
929 |
int ret = -1; |
930 |
BlockReopenQueueEntry *bs_entry, *next; |
931 |
Error *local_err = NULL;
|
932 |
|
933 |
assert(bs_queue != NULL);
|
934 |
|
935 |
bdrv_drain_all(); |
936 |
|
937 |
QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { |
938 |
if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
|
939 |
error_propagate(errp, local_err); |
940 |
goto cleanup;
|
941 |
} |
942 |
bs_entry->prepared = true;
|
943 |
} |
944 |
|
945 |
/* If we reach this point, we have success and just need to apply the
|
946 |
* changes
|
947 |
*/
|
948 |
QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { |
949 |
bdrv_reopen_commit(&bs_entry->state); |
950 |
} |
951 |
|
952 |
ret = 0;
|
953 |
|
954 |
cleanup:
|
955 |
QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { |
956 |
if (ret && bs_entry->prepared) {
|
957 |
bdrv_reopen_abort(&bs_entry->state); |
958 |
} |
959 |
g_free(bs_entry); |
960 |
} |
961 |
g_free(bs_queue); |
962 |
return ret;
|
963 |
} |
964 |
|
965 |
|
966 |
/* Reopen a single BlockDriverState with the specified flags. */
|
967 |
int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) |
968 |
{ |
969 |
int ret = -1; |
970 |
Error *local_err = NULL;
|
971 |
BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
|
972 |
|
973 |
ret = bdrv_reopen_multiple(queue, &local_err); |
974 |
if (local_err != NULL) { |
975 |
error_propagate(errp, local_err); |
976 |
} |
977 |
return ret;
|
978 |
} |
979 |
|
980 |
|
981 |
/*
|
982 |
* Prepares a BlockDriverState for reopen. All changes are staged in the
|
983 |
* 'opaque' field of the BDRVReopenState, which is used and allocated by
|
984 |
* the block driver layer .bdrv_reopen_prepare()
|
985 |
*
|
986 |
* bs is the BlockDriverState to reopen
|
987 |
* flags are the new open flags
|
988 |
* queue is the reopen queue
|
989 |
*
|
990 |
* Returns 0 on success, non-zero on error. On error errp will be set
|
991 |
* as well.
|
992 |
*
|
993 |
* On failure, bdrv_reopen_abort() will be called to clean up any data.
|
994 |
* It is the responsibility of the caller to then call the abort() or
|
995 |
* commit() for any other BDS that have been left in a prepare() state
|
996 |
*
|
997 |
*/
|
998 |
int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
|
999 |
Error **errp) |
1000 |
{ |
1001 |
int ret = -1; |
1002 |
Error *local_err = NULL;
|
1003 |
BlockDriver *drv; |
1004 |
|
1005 |
assert(reopen_state != NULL);
|
1006 |
assert(reopen_state->bs->drv != NULL);
|
1007 |
drv = reopen_state->bs->drv; |
1008 |
|
1009 |
/* if we are to stay read-only, do not allow permission change
|
1010 |
* to r/w */
|
1011 |
if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
|
1012 |
reopen_state->flags & BDRV_O_RDWR) { |
1013 |
error_set(errp, QERR_DEVICE_IS_READ_ONLY, |
1014 |
reopen_state->bs->device_name); |
1015 |
goto error;
|
1016 |
} |
1017 |
|
1018 |
|
1019 |
ret = bdrv_flush(reopen_state->bs); |
1020 |
if (ret) {
|
1021 |
error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
|
1022 |
strerror(-ret)); |
1023 |
goto error;
|
1024 |
} |
1025 |
|
1026 |
if (drv->bdrv_reopen_prepare) {
|
1027 |
ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
1028 |
if (ret) {
|
1029 |
if (local_err != NULL) { |
1030 |
error_propagate(errp, local_err); |
1031 |
} else {
|
1032 |
error_set(errp, QERR_OPEN_FILE_FAILED, |
1033 |
reopen_state->bs->filename); |
1034 |
} |
1035 |
goto error;
|
1036 |
} |
1037 |
} else {
|
1038 |
/* It is currently mandatory to have a bdrv_reopen_prepare()
|
1039 |
* handler for each supported drv. */
|
1040 |
error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, |
1041 |
drv->format_name, reopen_state->bs->device_name, |
1042 |
"reopening of file");
|
1043 |
ret = -1;
|
1044 |
goto error;
|
1045 |
} |
1046 |
|
1047 |
ret = 0;
|
1048 |
|
1049 |
error:
|
1050 |
return ret;
|
1051 |
} |
1052 |
|
1053 |
/*
|
1054 |
* Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
|
1055 |
* makes them final by swapping the staging BlockDriverState contents into
|
1056 |
* the active BlockDriverState contents.
|
1057 |
*/
|
1058 |
void bdrv_reopen_commit(BDRVReopenState *reopen_state)
|
1059 |
{ |
1060 |
BlockDriver *drv; |
1061 |
|
1062 |
assert(reopen_state != NULL);
|
1063 |
drv = reopen_state->bs->drv; |
1064 |
assert(drv != NULL);
|
1065 |
|
1066 |
/* If there are any driver level actions to take */
|
1067 |
if (drv->bdrv_reopen_commit) {
|
1068 |
drv->bdrv_reopen_commit(reopen_state); |
1069 |
} |
1070 |
|
1071 |
/* set BDS specific flags now */
|
1072 |
reopen_state->bs->open_flags = reopen_state->flags; |
1073 |
reopen_state->bs->enable_write_cache = !!(reopen_state->flags & |
1074 |
BDRV_O_CACHE_WB); |
1075 |
reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); |
1076 |
} |
1077 |
|
1078 |
/*
|
1079 |
* Abort the reopen, and delete and free the staged changes in
|
1080 |
* reopen_state
|
1081 |
*/
|
1082 |
void bdrv_reopen_abort(BDRVReopenState *reopen_state)
|
1083 |
{ |
1084 |
BlockDriver *drv; |
1085 |
|
1086 |
assert(reopen_state != NULL);
|
1087 |
drv = reopen_state->bs->drv; |
1088 |
assert(drv != NULL);
|
1089 |
|
1090 |
if (drv->bdrv_reopen_abort) {
|
1091 |
drv->bdrv_reopen_abort(reopen_state); |
1092 |
} |
1093 |
} |
1094 |
|
1095 |
|
1096 |
void bdrv_close(BlockDriverState *bs)
|
1097 |
{ |
1098 |
bdrv_flush(bs); |
1099 |
if (bs->drv) {
|
1100 |
if (bs->job) {
|
1101 |
block_job_cancel_sync(bs->job); |
1102 |
} |
1103 |
bdrv_drain_all(); |
1104 |
|
1105 |
if (bs == bs_snapshots) {
|
1106 |
bs_snapshots = NULL;
|
1107 |
} |
1108 |
if (bs->backing_hd) {
|
1109 |
bdrv_delete(bs->backing_hd); |
1110 |
bs->backing_hd = NULL;
|
1111 |
} |
1112 |
bs->drv->bdrv_close(bs); |
1113 |
g_free(bs->opaque); |
1114 |
#ifdef _WIN32
|
1115 |
if (bs->is_temporary) {
|
1116 |
unlink(bs->filename); |
1117 |
} |
1118 |
#endif
|
1119 |
bs->opaque = NULL;
|
1120 |
bs->drv = NULL;
|
1121 |
bs->copy_on_read = 0;
|
1122 |
bs->backing_file[0] = '\0'; |
1123 |
bs->backing_format[0] = '\0'; |
1124 |
bs->total_sectors = 0;
|
1125 |
bs->encrypted = 0;
|
1126 |
bs->valid_key = 0;
|
1127 |
bs->sg = 0;
|
1128 |
bs->growable = 0;
|
1129 |
|
1130 |
if (bs->file != NULL) { |
1131 |
bdrv_delete(bs->file); |
1132 |
bs->file = NULL;
|
1133 |
} |
1134 |
} |
1135 |
|
1136 |
bdrv_dev_change_media_cb(bs, false);
|
1137 |
|
1138 |
/*throttling disk I/O limits*/
|
1139 |
if (bs->io_limits_enabled) {
|
1140 |
bdrv_io_limits_disable(bs); |
1141 |
} |
1142 |
} |
1143 |
|
1144 |
void bdrv_close_all(void) |
1145 |
{ |
1146 |
BlockDriverState *bs; |
1147 |
|
1148 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
1149 |
bdrv_close(bs); |
1150 |
} |
1151 |
} |
1152 |
|
1153 |
/*
|
1154 |
* Wait for pending requests to complete across all BlockDriverStates
|
1155 |
*
|
1156 |
* This function does not flush data to disk, use bdrv_flush_all() for that
|
1157 |
* after calling this function.
|
1158 |
*
|
1159 |
* Note that completion of an asynchronous I/O operation can trigger any
|
1160 |
* number of other I/O operations on other devices---for example a coroutine
|
1161 |
* can be arbitrarily complex and a constant flow of I/O can come until the
|
1162 |
* coroutine is complete. Because of this, it is not possible to have a
|
1163 |
* function to drain a single device's I/O queue.
|
1164 |
*/
|
1165 |
void bdrv_drain_all(void) |
1166 |
{ |
1167 |
BlockDriverState *bs; |
1168 |
bool busy;
|
1169 |
|
1170 |
do {
|
1171 |
busy = qemu_aio_wait(); |
1172 |
|
1173 |
/* FIXME: We do not have timer support here, so this is effectively
|
1174 |
* a busy wait.
|
1175 |
*/
|
1176 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
1177 |
if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
|
1178 |
qemu_co_queue_restart_all(&bs->throttled_reqs); |
1179 |
busy = true;
|
1180 |
} |
1181 |
} |
1182 |
} while (busy);
|
1183 |
|
1184 |
/* If requests are still pending there is a bug somewhere */
|
1185 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
1186 |
assert(QLIST_EMPTY(&bs->tracked_requests)); |
1187 |
assert(qemu_co_queue_empty(&bs->throttled_reqs)); |
1188 |
} |
1189 |
} |
1190 |
|
1191 |
/* make a BlockDriverState anonymous by removing from bdrv_state list.
|
1192 |
Also, NULL terminate the device_name to prevent double remove */
|
1193 |
void bdrv_make_anon(BlockDriverState *bs)
|
1194 |
{ |
1195 |
if (bs->device_name[0] != '\0') { |
1196 |
QTAILQ_REMOVE(&bdrv_states, bs, list); |
1197 |
} |
1198 |
bs->device_name[0] = '\0'; |
1199 |
} |
1200 |
|
1201 |
static void bdrv_rebind(BlockDriverState *bs) |
1202 |
{ |
1203 |
if (bs->drv && bs->drv->bdrv_rebind) {
|
1204 |
bs->drv->bdrv_rebind(bs); |
1205 |
} |
1206 |
} |
1207 |
|
1208 |
static void bdrv_move_feature_fields(BlockDriverState *bs_dest, |
1209 |
BlockDriverState *bs_src) |
1210 |
{ |
1211 |
/* move some fields that need to stay attached to the device */
|
1212 |
bs_dest->open_flags = bs_src->open_flags; |
1213 |
|
1214 |
/* dev info */
|
1215 |
bs_dest->dev_ops = bs_src->dev_ops; |
1216 |
bs_dest->dev_opaque = bs_src->dev_opaque; |
1217 |
bs_dest->dev = bs_src->dev; |
1218 |
bs_dest->buffer_alignment = bs_src->buffer_alignment; |
1219 |
bs_dest->copy_on_read = bs_src->copy_on_read; |
1220 |
|
1221 |
bs_dest->enable_write_cache = bs_src->enable_write_cache; |
1222 |
|
1223 |
/* i/o timing parameters */
|
1224 |
bs_dest->slice_time = bs_src->slice_time; |
1225 |
bs_dest->slice_start = bs_src->slice_start; |
1226 |
bs_dest->slice_end = bs_src->slice_end; |
1227 |
bs_dest->io_limits = bs_src->io_limits; |
1228 |
bs_dest->io_base = bs_src->io_base; |
1229 |
bs_dest->throttled_reqs = bs_src->throttled_reqs; |
1230 |
bs_dest->block_timer = bs_src->block_timer; |
1231 |
bs_dest->io_limits_enabled = bs_src->io_limits_enabled; |
1232 |
|
1233 |
/* r/w error */
|
1234 |
bs_dest->on_read_error = bs_src->on_read_error; |
1235 |
bs_dest->on_write_error = bs_src->on_write_error; |
1236 |
|
1237 |
/* i/o status */
|
1238 |
bs_dest->iostatus_enabled = bs_src->iostatus_enabled; |
1239 |
bs_dest->iostatus = bs_src->iostatus; |
1240 |
|
1241 |
/* dirty bitmap */
|
1242 |
bs_dest->dirty_count = bs_src->dirty_count; |
1243 |
bs_dest->dirty_bitmap = bs_src->dirty_bitmap; |
1244 |
|
1245 |
/* job */
|
1246 |
bs_dest->in_use = bs_src->in_use; |
1247 |
bs_dest->job = bs_src->job; |
1248 |
|
1249 |
/* keep the same entry in bdrv_states */
|
1250 |
pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
|
1251 |
bs_src->device_name); |
1252 |
bs_dest->list = bs_src->list; |
1253 |
} |
1254 |
|
1255 |
/*
|
1256 |
* Swap bs contents for two image chains while they are live,
|
1257 |
* while keeping required fields on the BlockDriverState that is
|
1258 |
* actually attached to a device.
|
1259 |
*
|
1260 |
* This will modify the BlockDriverState fields, and swap contents
|
1261 |
* between bs_new and bs_old. Both bs_new and bs_old are modified.
|
1262 |
*
|
1263 |
* bs_new is required to be anonymous.
|
1264 |
*
|
1265 |
* This function does not create any image files.
|
1266 |
*/
|
1267 |
void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
|
1268 |
{ |
1269 |
BlockDriverState tmp; |
1270 |
|
1271 |
/* bs_new must be anonymous and shouldn't have anything fancy enabled */
|
1272 |
assert(bs_new->device_name[0] == '\0'); |
1273 |
assert(bs_new->dirty_bitmap == NULL);
|
1274 |
assert(bs_new->job == NULL);
|
1275 |
assert(bs_new->dev == NULL);
|
1276 |
assert(bs_new->in_use == 0);
|
1277 |
assert(bs_new->io_limits_enabled == false);
|
1278 |
assert(bs_new->block_timer == NULL);
|
1279 |
|
1280 |
tmp = *bs_new; |
1281 |
*bs_new = *bs_old; |
1282 |
*bs_old = tmp; |
1283 |
|
1284 |
/* there are some fields that should not be swapped, move them back */
|
1285 |
bdrv_move_feature_fields(&tmp, bs_old); |
1286 |
bdrv_move_feature_fields(bs_old, bs_new); |
1287 |
bdrv_move_feature_fields(bs_new, &tmp); |
1288 |
|
1289 |
/* bs_new shouldn't be in bdrv_states even after the swap! */
|
1290 |
assert(bs_new->device_name[0] == '\0'); |
1291 |
|
1292 |
/* Check a few fields that should remain attached to the device */
|
1293 |
assert(bs_new->dev == NULL);
|
1294 |
assert(bs_new->job == NULL);
|
1295 |
assert(bs_new->in_use == 0);
|
1296 |
assert(bs_new->io_limits_enabled == false);
|
1297 |
assert(bs_new->block_timer == NULL);
|
1298 |
|
1299 |
bdrv_rebind(bs_new); |
1300 |
bdrv_rebind(bs_old); |
1301 |
} |
1302 |
|
1303 |
/*
|
1304 |
* Add new bs contents at the top of an image chain while the chain is
|
1305 |
* live, while keeping required fields on the top layer.
|
1306 |
*
|
1307 |
* This will modify the BlockDriverState fields, and swap contents
|
1308 |
* between bs_new and bs_top. Both bs_new and bs_top are modified.
|
1309 |
*
|
1310 |
* bs_new is required to be anonymous.
|
1311 |
*
|
1312 |
* This function does not create any image files.
|
1313 |
*/
|
1314 |
void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
|
1315 |
{ |
1316 |
bdrv_swap(bs_new, bs_top); |
1317 |
|
1318 |
/* The contents of 'tmp' will become bs_top, as we are
|
1319 |
* swapping bs_new and bs_top contents. */
|
1320 |
bs_top->backing_hd = bs_new; |
1321 |
bs_top->open_flags &= ~BDRV_O_NO_BACKING; |
1322 |
pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
|
1323 |
bs_new->filename); |
1324 |
pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
|
1325 |
bs_new->drv ? bs_new->drv->format_name : "");
|
1326 |
} |
1327 |
|
1328 |
void bdrv_delete(BlockDriverState *bs)
|
1329 |
{ |
1330 |
assert(!bs->dev); |
1331 |
assert(!bs->job); |
1332 |
assert(!bs->in_use); |
1333 |
|
1334 |
/* remove from list, if necessary */
|
1335 |
bdrv_make_anon(bs); |
1336 |
|
1337 |
bdrv_close(bs); |
1338 |
|
1339 |
assert(bs != bs_snapshots); |
1340 |
g_free(bs); |
1341 |
} |
1342 |
|
1343 |
int bdrv_attach_dev(BlockDriverState *bs, void *dev) |
1344 |
/* TODO change to DeviceState *dev when all users are qdevified */
|
1345 |
{ |
1346 |
if (bs->dev) {
|
1347 |
return -EBUSY;
|
1348 |
} |
1349 |
bs->dev = dev; |
1350 |
bdrv_iostatus_reset(bs); |
1351 |
return 0; |
1352 |
} |
1353 |
|
1354 |
/* TODO qdevified devices don't use this, remove when devices are qdevified */
|
1355 |
void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev) |
1356 |
{ |
1357 |
if (bdrv_attach_dev(bs, dev) < 0) { |
1358 |
abort(); |
1359 |
} |
1360 |
} |
1361 |
|
1362 |
void bdrv_detach_dev(BlockDriverState *bs, void *dev) |
1363 |
/* TODO change to DeviceState *dev when all users are qdevified */
|
1364 |
{ |
1365 |
assert(bs->dev == dev); |
1366 |
bs->dev = NULL;
|
1367 |
bs->dev_ops = NULL;
|
1368 |
bs->dev_opaque = NULL;
|
1369 |
bs->buffer_alignment = 512;
|
1370 |
} |
1371 |
|
1372 |
/* TODO change to return DeviceState * when all users are qdevified */
|
1373 |
void *bdrv_get_attached_dev(BlockDriverState *bs)
|
1374 |
{ |
1375 |
return bs->dev;
|
1376 |
} |
1377 |
|
1378 |
void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops, |
1379 |
void *opaque)
|
1380 |
{ |
1381 |
bs->dev_ops = ops; |
1382 |
bs->dev_opaque = opaque; |
1383 |
if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
|
1384 |
bs_snapshots = NULL;
|
1385 |
} |
1386 |
} |
1387 |
|
1388 |
void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv, |
1389 |
BlockQMPEventAction action, int is_read)
|
1390 |
{ |
1391 |
QObject *data; |
1392 |
const char *action_str; |
1393 |
|
1394 |
switch (action) {
|
1395 |
case BDRV_ACTION_REPORT:
|
1396 |
action_str = "report";
|
1397 |
break;
|
1398 |
case BDRV_ACTION_IGNORE:
|
1399 |
action_str = "ignore";
|
1400 |
break;
|
1401 |
case BDRV_ACTION_STOP:
|
1402 |
action_str = "stop";
|
1403 |
break;
|
1404 |
default:
|
1405 |
abort(); |
1406 |
} |
1407 |
|
1408 |
data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
|
1409 |
bdrv->device_name, |
1410 |
action_str, |
1411 |
is_read ? "read" : "write"); |
1412 |
monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data); |
1413 |
|
1414 |
qobject_decref(data); |
1415 |
} |
1416 |
|
1417 |
static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected) |
1418 |
{ |
1419 |
QObject *data; |
1420 |
|
1421 |
data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
|
1422 |
bdrv_get_device_name(bs), ejected); |
1423 |
monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data); |
1424 |
|
1425 |
qobject_decref(data); |
1426 |
} |
1427 |
|
1428 |
static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load) |
1429 |
{ |
1430 |
if (bs->dev_ops && bs->dev_ops->change_media_cb) {
|
1431 |
bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
|
1432 |
bs->dev_ops->change_media_cb(bs->dev_opaque, load); |
1433 |
if (tray_was_closed) {
|
1434 |
/* tray open */
|
1435 |
bdrv_emit_qmp_eject_event(bs, true);
|
1436 |
} |
1437 |
if (load) {
|
1438 |
/* tray close */
|
1439 |
bdrv_emit_qmp_eject_event(bs, false);
|
1440 |
} |
1441 |
} |
1442 |
} |
1443 |
|
1444 |
bool bdrv_dev_has_removable_media(BlockDriverState *bs)
|
1445 |
{ |
1446 |
return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
|
1447 |
} |
1448 |
|
1449 |
void bdrv_dev_eject_request(BlockDriverState *bs, bool force) |
1450 |
{ |
1451 |
if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
|
1452 |
bs->dev_ops->eject_request_cb(bs->dev_opaque, force); |
1453 |
} |
1454 |
} |
1455 |
|
1456 |
bool bdrv_dev_is_tray_open(BlockDriverState *bs)
|
1457 |
{ |
1458 |
if (bs->dev_ops && bs->dev_ops->is_tray_open) {
|
1459 |
return bs->dev_ops->is_tray_open(bs->dev_opaque);
|
1460 |
} |
1461 |
return false; |
1462 |
} |
1463 |
|
1464 |
static void bdrv_dev_resize_cb(BlockDriverState *bs) |
1465 |
{ |
1466 |
if (bs->dev_ops && bs->dev_ops->resize_cb) {
|
1467 |
bs->dev_ops->resize_cb(bs->dev_opaque); |
1468 |
} |
1469 |
} |
1470 |
|
1471 |
bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
|
1472 |
{ |
1473 |
if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
|
1474 |
return bs->dev_ops->is_medium_locked(bs->dev_opaque);
|
1475 |
} |
1476 |
return false; |
1477 |
} |
1478 |
|
1479 |
/*
|
1480 |
* Run consistency checks on an image
|
1481 |
*
|
1482 |
* Returns 0 if the check could be completed (it doesn't mean that the image is
|
1483 |
* free of errors) or -errno when an internal error occurred. The results of the
|
1484 |
* check are stored in res.
|
1485 |
*/
|
1486 |
int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
|
1487 |
{ |
1488 |
if (bs->drv->bdrv_check == NULL) { |
1489 |
return -ENOTSUP;
|
1490 |
} |
1491 |
|
1492 |
memset(res, 0, sizeof(*res)); |
1493 |
return bs->drv->bdrv_check(bs, res, fix);
|
1494 |
} |
1495 |
|
1496 |
#define COMMIT_BUF_SECTORS 2048 |
1497 |
|
1498 |
/* commit COW file into the raw image */
|
1499 |
int bdrv_commit(BlockDriverState *bs)
|
1500 |
{ |
1501 |
BlockDriver *drv = bs->drv; |
1502 |
int64_t sector, total_sectors; |
1503 |
int n, ro, open_flags;
|
1504 |
int ret = 0; |
1505 |
uint8_t *buf; |
1506 |
char filename[1024]; |
1507 |
|
1508 |
if (!drv)
|
1509 |
return -ENOMEDIUM;
|
1510 |
|
1511 |
if (!bs->backing_hd) {
|
1512 |
return -ENOTSUP;
|
1513 |
} |
1514 |
|
1515 |
if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
|
1516 |
return -EBUSY;
|
1517 |
} |
1518 |
|
1519 |
ro = bs->backing_hd->read_only; |
1520 |
strncpy(filename, bs->backing_hd->filename, sizeof(filename));
|
1521 |
open_flags = bs->backing_hd->open_flags; |
1522 |
|
1523 |
if (ro) {
|
1524 |
if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) { |
1525 |
return -EACCES;
|
1526 |
} |
1527 |
} |
1528 |
|
1529 |
total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; |
1530 |
buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE); |
1531 |
|
1532 |
for (sector = 0; sector < total_sectors; sector += n) { |
1533 |
if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
|
1534 |
|
1535 |
if (bdrv_read(bs, sector, buf, n) != 0) { |
1536 |
ret = -EIO; |
1537 |
goto ro_cleanup;
|
1538 |
} |
1539 |
|
1540 |
if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) { |
1541 |
ret = -EIO; |
1542 |
goto ro_cleanup;
|
1543 |
} |
1544 |
} |
1545 |
} |
1546 |
|
1547 |
if (drv->bdrv_make_empty) {
|
1548 |
ret = drv->bdrv_make_empty(bs); |
1549 |
bdrv_flush(bs); |
1550 |
} |
1551 |
|
1552 |
/*
|
1553 |
* Make sure all data we wrote to the backing device is actually
|
1554 |
* stable on disk.
|
1555 |
*/
|
1556 |
if (bs->backing_hd)
|
1557 |
bdrv_flush(bs->backing_hd); |
1558 |
|
1559 |
ro_cleanup:
|
1560 |
g_free(buf); |
1561 |
|
1562 |
if (ro) {
|
1563 |
/* ignoring error return here */
|
1564 |
bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
|
1565 |
} |
1566 |
|
1567 |
return ret;
|
1568 |
} |
1569 |
|
1570 |
int bdrv_commit_all(void) |
1571 |
{ |
1572 |
BlockDriverState *bs; |
1573 |
|
1574 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
1575 |
int ret = bdrv_commit(bs);
|
1576 |
if (ret < 0) { |
1577 |
return ret;
|
1578 |
} |
1579 |
} |
1580 |
return 0; |
1581 |
} |
1582 |
|
1583 |
struct BdrvTrackedRequest {
|
1584 |
BlockDriverState *bs; |
1585 |
int64_t sector_num; |
1586 |
int nb_sectors;
|
1587 |
bool is_write;
|
1588 |
QLIST_ENTRY(BdrvTrackedRequest) list; |
1589 |
Coroutine *co; /* owner, used for deadlock detection */
|
1590 |
CoQueue wait_queue; /* coroutines blocked on this request */
|
1591 |
}; |
1592 |
|
1593 |
/**
|
1594 |
* Remove an active request from the tracked requests list
|
1595 |
*
|
1596 |
* This function should be called when a tracked request is completing.
|
1597 |
*/
|
1598 |
static void tracked_request_end(BdrvTrackedRequest *req) |
1599 |
{ |
1600 |
QLIST_REMOVE(req, list); |
1601 |
qemu_co_queue_restart_all(&req->wait_queue); |
1602 |
} |
1603 |
|
1604 |
/**
|
1605 |
* Add an active request to the tracked requests list
|
1606 |
*/
|
1607 |
static void tracked_request_begin(BdrvTrackedRequest *req, |
1608 |
BlockDriverState *bs, |
1609 |
int64_t sector_num, |
1610 |
int nb_sectors, bool is_write) |
1611 |
{ |
1612 |
*req = (BdrvTrackedRequest){ |
1613 |
.bs = bs, |
1614 |
.sector_num = sector_num, |
1615 |
.nb_sectors = nb_sectors, |
1616 |
.is_write = is_write, |
1617 |
.co = qemu_coroutine_self(), |
1618 |
}; |
1619 |
|
1620 |
qemu_co_queue_init(&req->wait_queue); |
1621 |
|
1622 |
QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
1623 |
} |
1624 |
|
1625 |
/**
|
1626 |
* Round a region to cluster boundaries
|
1627 |
*/
|
1628 |
static void round_to_clusters(BlockDriverState *bs, |
1629 |
int64_t sector_num, int nb_sectors,
|
1630 |
int64_t *cluster_sector_num, |
1631 |
int *cluster_nb_sectors)
|
1632 |
{ |
1633 |
BlockDriverInfo bdi; |
1634 |
|
1635 |
if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { |
1636 |
*cluster_sector_num = sector_num; |
1637 |
*cluster_nb_sectors = nb_sectors; |
1638 |
} else {
|
1639 |
int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; |
1640 |
*cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); |
1641 |
*cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + |
1642 |
nb_sectors, c); |
1643 |
} |
1644 |
} |
1645 |
|
1646 |
static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
1647 |
int64_t sector_num, int nb_sectors) {
|
1648 |
/* aaaa bbbb */
|
1649 |
if (sector_num >= req->sector_num + req->nb_sectors) {
|
1650 |
return false; |
1651 |
} |
1652 |
/* bbbb aaaa */
|
1653 |
if (req->sector_num >= sector_num + nb_sectors) {
|
1654 |
return false; |
1655 |
} |
1656 |
return true; |
1657 |
} |
1658 |
|
1659 |
static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs, |
1660 |
int64_t sector_num, int nb_sectors)
|
1661 |
{ |
1662 |
BdrvTrackedRequest *req; |
1663 |
int64_t cluster_sector_num; |
1664 |
int cluster_nb_sectors;
|
1665 |
bool retry;
|
1666 |
|
1667 |
/* If we touch the same cluster it counts as an overlap. This guarantees
|
1668 |
* that allocating writes will be serialized and not race with each other
|
1669 |
* for the same cluster. For example, in copy-on-read it ensures that the
|
1670 |
* CoR read and write operations are atomic and guest writes cannot
|
1671 |
* interleave between them.
|
1672 |
*/
|
1673 |
round_to_clusters(bs, sector_num, nb_sectors, |
1674 |
&cluster_sector_num, &cluster_nb_sectors); |
1675 |
|
1676 |
do {
|
1677 |
retry = false;
|
1678 |
QLIST_FOREACH(req, &bs->tracked_requests, list) { |
1679 |
if (tracked_request_overlaps(req, cluster_sector_num,
|
1680 |
cluster_nb_sectors)) { |
1681 |
/* Hitting this means there was a reentrant request, for
|
1682 |
* example, a block driver issuing nested requests. This must
|
1683 |
* never happen since it means deadlock.
|
1684 |
*/
|
1685 |
assert(qemu_coroutine_self() != req->co); |
1686 |
|
1687 |
qemu_co_queue_wait(&req->wait_queue); |
1688 |
retry = true;
|
1689 |
break;
|
1690 |
} |
1691 |
} |
1692 |
} while (retry);
|
1693 |
} |
1694 |
|
1695 |
/*
|
1696 |
* Return values:
|
1697 |
* 0 - success
|
1698 |
* -EINVAL - backing format specified, but no file
|
1699 |
* -ENOSPC - can't update the backing file because no space is left in the
|
1700 |
* image file header
|
1701 |
* -ENOTSUP - format driver doesn't support changing the backing file
|
1702 |
*/
|
1703 |
int bdrv_change_backing_file(BlockDriverState *bs,
|
1704 |
const char *backing_file, const char *backing_fmt) |
1705 |
{ |
1706 |
BlockDriver *drv = bs->drv; |
1707 |
int ret;
|
1708 |
|
1709 |
/* Backing file format doesn't make sense without a backing file */
|
1710 |
if (backing_fmt && !backing_file) {
|
1711 |
return -EINVAL;
|
1712 |
} |
1713 |
|
1714 |
if (drv->bdrv_change_backing_file != NULL) { |
1715 |
ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
1716 |
} else {
|
1717 |
ret = -ENOTSUP; |
1718 |
} |
1719 |
|
1720 |
if (ret == 0) { |
1721 |
pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); |
1722 |
pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); |
1723 |
} |
1724 |
return ret;
|
1725 |
} |
1726 |
|
1727 |
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, |
1728 |
size_t size) |
1729 |
{ |
1730 |
int64_t len; |
1731 |
|
1732 |
if (!bdrv_is_inserted(bs))
|
1733 |
return -ENOMEDIUM;
|
1734 |
|
1735 |
if (bs->growable)
|
1736 |
return 0; |
1737 |
|
1738 |
len = bdrv_getlength(bs); |
1739 |
|
1740 |
if (offset < 0) |
1741 |
return -EIO;
|
1742 |
|
1743 |
if ((offset > len) || (len - offset < size))
|
1744 |
return -EIO;
|
1745 |
|
1746 |
return 0; |
1747 |
} |
1748 |
|
1749 |
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, |
1750 |
int nb_sectors)
|
1751 |
{ |
1752 |
return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
|
1753 |
nb_sectors * BDRV_SECTOR_SIZE); |
1754 |
} |
1755 |
|
1756 |
typedef struct RwCo { |
1757 |
BlockDriverState *bs; |
1758 |
int64_t sector_num; |
1759 |
int nb_sectors;
|
1760 |
QEMUIOVector *qiov; |
1761 |
bool is_write;
|
1762 |
int ret;
|
1763 |
} RwCo; |
1764 |
|
1765 |
static void coroutine_fn bdrv_rw_co_entry(void *opaque) |
1766 |
{ |
1767 |
RwCo *rwco = opaque; |
1768 |
|
1769 |
if (!rwco->is_write) {
|
1770 |
rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num, |
1771 |
rwco->nb_sectors, rwco->qiov, 0);
|
1772 |
} else {
|
1773 |
rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num, |
1774 |
rwco->nb_sectors, rwco->qiov, 0);
|
1775 |
} |
1776 |
} |
1777 |
|
1778 |
/*
|
1779 |
* Process a synchronous request using coroutines
|
1780 |
*/
|
1781 |
static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, |
1782 |
int nb_sectors, bool is_write) |
1783 |
{ |
1784 |
QEMUIOVector qiov; |
1785 |
struct iovec iov = {
|
1786 |
.iov_base = (void *)buf,
|
1787 |
.iov_len = nb_sectors * BDRV_SECTOR_SIZE, |
1788 |
}; |
1789 |
Coroutine *co; |
1790 |
RwCo rwco = { |
1791 |
.bs = bs, |
1792 |
.sector_num = sector_num, |
1793 |
.nb_sectors = nb_sectors, |
1794 |
.qiov = &qiov, |
1795 |
.is_write = is_write, |
1796 |
.ret = NOT_DONE, |
1797 |
}; |
1798 |
|
1799 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
1800 |
|
1801 |
/**
|
1802 |
* In sync call context, when the vcpu is blocked, this throttling timer
|
1803 |
* will not fire; so the I/O throttling function has to be disabled here
|
1804 |
* if it has been enabled.
|
1805 |
*/
|
1806 |
if (bs->io_limits_enabled) {
|
1807 |
fprintf(stderr, "Disabling I/O throttling on '%s' due "
|
1808 |
"to synchronous I/O.\n", bdrv_get_device_name(bs));
|
1809 |
bdrv_io_limits_disable(bs); |
1810 |
} |
1811 |
|
1812 |
if (qemu_in_coroutine()) {
|
1813 |
/* Fast-path if already in coroutine context */
|
1814 |
bdrv_rw_co_entry(&rwco); |
1815 |
} else {
|
1816 |
co = qemu_coroutine_create(bdrv_rw_co_entry); |
1817 |
qemu_coroutine_enter(co, &rwco); |
1818 |
while (rwco.ret == NOT_DONE) {
|
1819 |
qemu_aio_wait(); |
1820 |
} |
1821 |
} |
1822 |
return rwco.ret;
|
1823 |
} |
1824 |
|
1825 |
/* return < 0 if error. See bdrv_write() for the return codes */
|
1826 |
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
|
1827 |
uint8_t *buf, int nb_sectors)
|
1828 |
{ |
1829 |
return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false); |
1830 |
} |
1831 |
|
1832 |
/* Just like bdrv_read(), but with I/O throttling temporarily disabled */
|
1833 |
int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
|
1834 |
uint8_t *buf, int nb_sectors)
|
1835 |
{ |
1836 |
bool enabled;
|
1837 |
int ret;
|
1838 |
|
1839 |
enabled = bs->io_limits_enabled; |
1840 |
bs->io_limits_enabled = false;
|
1841 |
ret = bdrv_read(bs, 0, buf, 1); |
1842 |
bs->io_limits_enabled = enabled; |
1843 |
return ret;
|
1844 |
} |
1845 |
|
1846 |
#define BITS_PER_LONG (sizeof(unsigned long) * 8) |
1847 |
|
1848 |
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num, |
1849 |
int nb_sectors, int dirty) |
1850 |
{ |
1851 |
int64_t start, end; |
1852 |
unsigned long val, idx, bit; |
1853 |
|
1854 |
start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; |
1855 |
end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
|
1856 |
|
1857 |
for (; start <= end; start++) {
|
1858 |
idx = start / BITS_PER_LONG; |
1859 |
bit = start % BITS_PER_LONG; |
1860 |
val = bs->dirty_bitmap[idx]; |
1861 |
if (dirty) {
|
1862 |
if (!(val & (1UL << bit))) { |
1863 |
bs->dirty_count++; |
1864 |
val |= 1UL << bit;
|
1865 |
} |
1866 |
} else {
|
1867 |
if (val & (1UL << bit)) { |
1868 |
bs->dirty_count--; |
1869 |
val &= ~(1UL << bit);
|
1870 |
} |
1871 |
} |
1872 |
bs->dirty_bitmap[idx] = val; |
1873 |
} |
1874 |
} |
1875 |
|
1876 |
/* Return < 0 if error. Important errors are:
|
1877 |
-EIO generic I/O error (may happen for all errors)
|
1878 |
-ENOMEDIUM No media inserted.
|
1879 |
-EINVAL Invalid sector number or nb_sectors
|
1880 |
-EACCES Trying to write a read-only device
|
1881 |
*/
|
1882 |
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
|
1883 |
const uint8_t *buf, int nb_sectors) |
1884 |
{ |
1885 |
return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true); |
1886 |
} |
1887 |
|
1888 |
int bdrv_pread(BlockDriverState *bs, int64_t offset,
|
1889 |
void *buf, int count1) |
1890 |
{ |
1891 |
uint8_t tmp_buf[BDRV_SECTOR_SIZE]; |
1892 |
int len, nb_sectors, count;
|
1893 |
int64_t sector_num; |
1894 |
int ret;
|
1895 |
|
1896 |
count = count1; |
1897 |
/* first read to align to sector start */
|
1898 |
len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
|
1899 |
if (len > count)
|
1900 |
len = count; |
1901 |
sector_num = offset >> BDRV_SECTOR_BITS; |
1902 |
if (len > 0) { |
1903 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1904 |
return ret;
|
1905 |
memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
|
1906 |
count -= len; |
1907 |
if (count == 0) |
1908 |
return count1;
|
1909 |
sector_num++; |
1910 |
buf += len; |
1911 |
} |
1912 |
|
1913 |
/* read the sectors "in place" */
|
1914 |
nb_sectors = count >> BDRV_SECTOR_BITS; |
1915 |
if (nb_sectors > 0) { |
1916 |
if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0) |
1917 |
return ret;
|
1918 |
sector_num += nb_sectors; |
1919 |
len = nb_sectors << BDRV_SECTOR_BITS; |
1920 |
buf += len; |
1921 |
count -= len; |
1922 |
} |
1923 |
|
1924 |
/* add data from the last sector */
|
1925 |
if (count > 0) { |
1926 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1927 |
return ret;
|
1928 |
memcpy(buf, tmp_buf, count); |
1929 |
} |
1930 |
return count1;
|
1931 |
} |
1932 |
|
1933 |
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
|
1934 |
const void *buf, int count1) |
1935 |
{ |
1936 |
uint8_t tmp_buf[BDRV_SECTOR_SIZE]; |
1937 |
int len, nb_sectors, count;
|
1938 |
int64_t sector_num; |
1939 |
int ret;
|
1940 |
|
1941 |
count = count1; |
1942 |
/* first write to align to sector start */
|
1943 |
len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
|
1944 |
if (len > count)
|
1945 |
len = count; |
1946 |
sector_num = offset >> BDRV_SECTOR_BITS; |
1947 |
if (len > 0) { |
1948 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1949 |
return ret;
|
1950 |
memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
|
1951 |
if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) |
1952 |
return ret;
|
1953 |
count -= len; |
1954 |
if (count == 0) |
1955 |
return count1;
|
1956 |
sector_num++; |
1957 |
buf += len; |
1958 |
} |
1959 |
|
1960 |
/* write the sectors "in place" */
|
1961 |
nb_sectors = count >> BDRV_SECTOR_BITS; |
1962 |
if (nb_sectors > 0) { |
1963 |
if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0) |
1964 |
return ret;
|
1965 |
sector_num += nb_sectors; |
1966 |
len = nb_sectors << BDRV_SECTOR_BITS; |
1967 |
buf += len; |
1968 |
count -= len; |
1969 |
} |
1970 |
|
1971 |
/* add data from the last sector */
|
1972 |
if (count > 0) { |
1973 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1974 |
return ret;
|
1975 |
memcpy(tmp_buf, buf, count); |
1976 |
if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) |
1977 |
return ret;
|
1978 |
} |
1979 |
return count1;
|
1980 |
} |
1981 |
|
1982 |
/*
|
1983 |
* Writes to the file and ensures that no writes are reordered across this
|
1984 |
* request (acts as a barrier)
|
1985 |
*
|
1986 |
* Returns 0 on success, -errno in error cases.
|
1987 |
*/
|
1988 |
int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
|
1989 |
const void *buf, int count) |
1990 |
{ |
1991 |
int ret;
|
1992 |
|
1993 |
ret = bdrv_pwrite(bs, offset, buf, count); |
1994 |
if (ret < 0) { |
1995 |
return ret;
|
1996 |
} |
1997 |
|
1998 |
/* No flush needed for cache modes that already do it */
|
1999 |
if (bs->enable_write_cache) {
|
2000 |
bdrv_flush(bs); |
2001 |
} |
2002 |
|
2003 |
return 0; |
2004 |
} |
2005 |
|
2006 |
static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, |
2007 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
|
2008 |
{ |
2009 |
/* Perform I/O through a temporary buffer so that users who scribble over
|
2010 |
* their read buffer while the operation is in progress do not end up
|
2011 |
* modifying the image file. This is critical for zero-copy guest I/O
|
2012 |
* where anything might happen inside guest memory.
|
2013 |
*/
|
2014 |
void *bounce_buffer;
|
2015 |
|
2016 |
BlockDriver *drv = bs->drv; |
2017 |
struct iovec iov;
|
2018 |
QEMUIOVector bounce_qiov; |
2019 |
int64_t cluster_sector_num; |
2020 |
int cluster_nb_sectors;
|
2021 |
size_t skip_bytes; |
2022 |
int ret;
|
2023 |
|
2024 |
/* Cover entire cluster so no additional backing file I/O is required when
|
2025 |
* allocating cluster in the image file.
|
2026 |
*/
|
2027 |
round_to_clusters(bs, sector_num, nb_sectors, |
2028 |
&cluster_sector_num, &cluster_nb_sectors); |
2029 |
|
2030 |
trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, |
2031 |
cluster_sector_num, cluster_nb_sectors); |
2032 |
|
2033 |
iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; |
2034 |
iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len); |
2035 |
qemu_iovec_init_external(&bounce_qiov, &iov, 1);
|
2036 |
|
2037 |
ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, |
2038 |
&bounce_qiov); |
2039 |
if (ret < 0) { |
2040 |
goto err;
|
2041 |
} |
2042 |
|
2043 |
if (drv->bdrv_co_write_zeroes &&
|
2044 |
buffer_is_zero(bounce_buffer, iov.iov_len)) { |
2045 |
ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, |
2046 |
cluster_nb_sectors); |
2047 |
} else {
|
2048 |
/* This does not change the data on the disk, it is not necessary
|
2049 |
* to flush even in cache=writethrough mode.
|
2050 |
*/
|
2051 |
ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, |
2052 |
&bounce_qiov); |
2053 |
} |
2054 |
|
2055 |
if (ret < 0) { |
2056 |
/* It might be okay to ignore write errors for guest requests. If this
|
2057 |
* is a deliberate copy-on-read then we don't want to ignore the error.
|
2058 |
* Simply report it in all cases.
|
2059 |
*/
|
2060 |
goto err;
|
2061 |
} |
2062 |
|
2063 |
skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; |
2064 |
qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
|
2065 |
nb_sectors * BDRV_SECTOR_SIZE); |
2066 |
|
2067 |
err:
|
2068 |
qemu_vfree(bounce_buffer); |
2069 |
return ret;
|
2070 |
} |
2071 |
|
2072 |
/*
|
2073 |
* Handle a read request in coroutine context
|
2074 |
*/
|
2075 |
static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, |
2076 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
2077 |
BdrvRequestFlags flags) |
2078 |
{ |
2079 |
BlockDriver *drv = bs->drv; |
2080 |
BdrvTrackedRequest req; |
2081 |
int ret;
|
2082 |
|
2083 |
if (!drv) {
|
2084 |
return -ENOMEDIUM;
|
2085 |
} |
2086 |
if (bdrv_check_request(bs, sector_num, nb_sectors)) {
|
2087 |
return -EIO;
|
2088 |
} |
2089 |
|
2090 |
/* throttling disk read I/O */
|
2091 |
if (bs->io_limits_enabled) {
|
2092 |
bdrv_io_limits_intercept(bs, false, nb_sectors);
|
2093 |
} |
2094 |
|
2095 |
if (bs->copy_on_read) {
|
2096 |
flags |= BDRV_REQ_COPY_ON_READ; |
2097 |
} |
2098 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
2099 |
bs->copy_on_read_in_flight++; |
2100 |
} |
2101 |
|
2102 |
if (bs->copy_on_read_in_flight) {
|
2103 |
wait_for_overlapping_requests(bs, sector_num, nb_sectors); |
2104 |
} |
2105 |
|
2106 |
tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
|
2107 |
|
2108 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
2109 |
int pnum;
|
2110 |
|
2111 |
ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum); |
2112 |
if (ret < 0) { |
2113 |
goto out;
|
2114 |
} |
2115 |
|
2116 |
if (!ret || pnum != nb_sectors) {
|
2117 |
ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); |
2118 |
goto out;
|
2119 |
} |
2120 |
} |
2121 |
|
2122 |
ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
2123 |
|
2124 |
out:
|
2125 |
tracked_request_end(&req); |
2126 |
|
2127 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
2128 |
bs->copy_on_read_in_flight--; |
2129 |
} |
2130 |
|
2131 |
return ret;
|
2132 |
} |
2133 |
|
2134 |
int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
|
2135 |
int nb_sectors, QEMUIOVector *qiov)
|
2136 |
{ |
2137 |
trace_bdrv_co_readv(bs, sector_num, nb_sectors); |
2138 |
|
2139 |
return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); |
2140 |
} |
2141 |
|
2142 |
int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
|
2143 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
|
2144 |
{ |
2145 |
trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); |
2146 |
|
2147 |
return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
|
2148 |
BDRV_REQ_COPY_ON_READ); |
2149 |
} |
2150 |
|
2151 |
static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, |
2152 |
int64_t sector_num, int nb_sectors)
|
2153 |
{ |
2154 |
BlockDriver *drv = bs->drv; |
2155 |
QEMUIOVector qiov; |
2156 |
struct iovec iov;
|
2157 |
int ret;
|
2158 |
|
2159 |
/* TODO Emulate only part of misaligned requests instead of letting block
|
2160 |
* drivers return -ENOTSUP and emulate everything */
|
2161 |
|
2162 |
/* First try the efficient write zeroes operation */
|
2163 |
if (drv->bdrv_co_write_zeroes) {
|
2164 |
ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors); |
2165 |
if (ret != -ENOTSUP) {
|
2166 |
return ret;
|
2167 |
} |
2168 |
} |
2169 |
|
2170 |
/* Fall back to bounce buffer if write zeroes is unsupported */
|
2171 |
iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE; |
2172 |
iov.iov_base = qemu_blockalign(bs, iov.iov_len); |
2173 |
memset(iov.iov_base, 0, iov.iov_len);
|
2174 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
2175 |
|
2176 |
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov); |
2177 |
|
2178 |
qemu_vfree(iov.iov_base); |
2179 |
return ret;
|
2180 |
} |
2181 |
|
2182 |
/*
|
2183 |
* Handle a write request in coroutine context
|
2184 |
*/
|
2185 |
static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, |
2186 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
2187 |
BdrvRequestFlags flags) |
2188 |
{ |
2189 |
BlockDriver *drv = bs->drv; |
2190 |
BdrvTrackedRequest req; |
2191 |
int ret;
|
2192 |
|
2193 |
if (!bs->drv) {
|
2194 |
return -ENOMEDIUM;
|
2195 |
} |
2196 |
if (bs->read_only) {
|
2197 |
return -EACCES;
|
2198 |
} |
2199 |
if (bdrv_check_request(bs, sector_num, nb_sectors)) {
|
2200 |
return -EIO;
|
2201 |
} |
2202 |
|
2203 |
/* throttling disk write I/O */
|
2204 |
if (bs->io_limits_enabled) {
|
2205 |
bdrv_io_limits_intercept(bs, true, nb_sectors);
|
2206 |
} |
2207 |
|
2208 |
if (bs->copy_on_read_in_flight) {
|
2209 |
wait_for_overlapping_requests(bs, sector_num, nb_sectors); |
2210 |
} |
2211 |
|
2212 |
tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
|
2213 |
|
2214 |
if (flags & BDRV_REQ_ZERO_WRITE) {
|
2215 |
ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors); |
2216 |
} else {
|
2217 |
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
2218 |
} |
2219 |
|
2220 |
if (ret == 0 && !bs->enable_write_cache) { |
2221 |
ret = bdrv_co_flush(bs); |
2222 |
} |
2223 |
|
2224 |
if (bs->dirty_bitmap) {
|
2225 |
set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
|
2226 |
} |
2227 |
|
2228 |
if (bs->wr_highest_sector < sector_num + nb_sectors - 1) { |
2229 |
bs->wr_highest_sector = sector_num + nb_sectors - 1;
|
2230 |
} |
2231 |
|
2232 |
tracked_request_end(&req); |
2233 |
|
2234 |
return ret;
|
2235 |
} |
2236 |
|
2237 |
int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
|
2238 |
int nb_sectors, QEMUIOVector *qiov)
|
2239 |
{ |
2240 |
trace_bdrv_co_writev(bs, sector_num, nb_sectors); |
2241 |
|
2242 |
return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); |
2243 |
} |
2244 |
|
2245 |
int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
|
2246 |
int64_t sector_num, int nb_sectors)
|
2247 |
{ |
2248 |
trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors); |
2249 |
|
2250 |
return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, |
2251 |
BDRV_REQ_ZERO_WRITE); |
2252 |
} |
2253 |
|
2254 |
/**
|
2255 |
* Truncate file to 'offset' bytes (needed only for file protocols)
|
2256 |
*/
|
2257 |
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
|
2258 |
{ |
2259 |
BlockDriver *drv = bs->drv; |
2260 |
int ret;
|
2261 |
if (!drv)
|
2262 |
return -ENOMEDIUM;
|
2263 |
if (!drv->bdrv_truncate)
|
2264 |
return -ENOTSUP;
|
2265 |
if (bs->read_only)
|
2266 |
return -EACCES;
|
2267 |
if (bdrv_in_use(bs))
|
2268 |
return -EBUSY;
|
2269 |
ret = drv->bdrv_truncate(bs, offset); |
2270 |
if (ret == 0) { |
2271 |
ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
2272 |
bdrv_dev_resize_cb(bs); |
2273 |
} |
2274 |
return ret;
|
2275 |
} |
2276 |
|
2277 |
/**
|
2278 |
* Length of a allocated file in bytes. Sparse files are counted by actual
|
2279 |
* allocated space. Return < 0 if error or unknown.
|
2280 |
*/
|
2281 |
int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) |
2282 |
{ |
2283 |
BlockDriver *drv = bs->drv; |
2284 |
if (!drv) {
|
2285 |
return -ENOMEDIUM;
|
2286 |
} |
2287 |
if (drv->bdrv_get_allocated_file_size) {
|
2288 |
return drv->bdrv_get_allocated_file_size(bs);
|
2289 |
} |
2290 |
if (bs->file) {
|
2291 |
return bdrv_get_allocated_file_size(bs->file);
|
2292 |
} |
2293 |
return -ENOTSUP;
|
2294 |
} |
2295 |
|
2296 |
/**
|
2297 |
* Length of a file in bytes. Return < 0 if error or unknown.
|
2298 |
*/
|
2299 |
int64_t bdrv_getlength(BlockDriverState *bs) |
2300 |
{ |
2301 |
BlockDriver *drv = bs->drv; |
2302 |
if (!drv)
|
2303 |
return -ENOMEDIUM;
|
2304 |
|
2305 |
if (bs->growable || bdrv_dev_has_removable_media(bs)) {
|
2306 |
if (drv->bdrv_getlength) {
|
2307 |
return drv->bdrv_getlength(bs);
|
2308 |
} |
2309 |
} |
2310 |
return bs->total_sectors * BDRV_SECTOR_SIZE;
|
2311 |
} |
2312 |
|
2313 |
/* return 0 as number of sectors if no device present or error */
|
2314 |
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
|
2315 |
{ |
2316 |
int64_t length; |
2317 |
length = bdrv_getlength(bs); |
2318 |
if (length < 0) |
2319 |
length = 0;
|
2320 |
else
|
2321 |
length = length >> BDRV_SECTOR_BITS; |
2322 |
*nb_sectors_ptr = length; |
2323 |
} |
2324 |
|
2325 |
/* throttling disk io limits */
|
2326 |
void bdrv_set_io_limits(BlockDriverState *bs,
|
2327 |
BlockIOLimit *io_limits) |
2328 |
{ |
2329 |
bs->io_limits = *io_limits; |
2330 |
bs->io_limits_enabled = bdrv_io_limits_enabled(bs); |
2331 |
} |
2332 |
|
2333 |
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
|
2334 |
BlockErrorAction on_write_error) |
2335 |
{ |
2336 |
bs->on_read_error = on_read_error; |
2337 |
bs->on_write_error = on_write_error; |
2338 |
} |
2339 |
|
2340 |
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
|
2341 |
{ |
2342 |
return is_read ? bs->on_read_error : bs->on_write_error;
|
2343 |
} |
2344 |
|
2345 |
int bdrv_is_read_only(BlockDriverState *bs)
|
2346 |
{ |
2347 |
return bs->read_only;
|
2348 |
} |
2349 |
|
2350 |
int bdrv_is_sg(BlockDriverState *bs)
|
2351 |
{ |
2352 |
return bs->sg;
|
2353 |
} |
2354 |
|
2355 |
int bdrv_enable_write_cache(BlockDriverState *bs)
|
2356 |
{ |
2357 |
return bs->enable_write_cache;
|
2358 |
} |
2359 |
|
2360 |
void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce) |
2361 |
{ |
2362 |
bs->enable_write_cache = wce; |
2363 |
|
2364 |
/* so a reopen() will preserve wce */
|
2365 |
if (wce) {
|
2366 |
bs->open_flags |= BDRV_O_CACHE_WB; |
2367 |
} else {
|
2368 |
bs->open_flags &= ~BDRV_O_CACHE_WB; |
2369 |
} |
2370 |
} |
2371 |
|
2372 |
int bdrv_is_encrypted(BlockDriverState *bs)
|
2373 |
{ |
2374 |
if (bs->backing_hd && bs->backing_hd->encrypted)
|
2375 |
return 1; |
2376 |
return bs->encrypted;
|
2377 |
} |
2378 |
|
2379 |
int bdrv_key_required(BlockDriverState *bs)
|
2380 |
{ |
2381 |
BlockDriverState *backing_hd = bs->backing_hd; |
2382 |
|
2383 |
if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
|
2384 |
return 1; |
2385 |
return (bs->encrypted && !bs->valid_key);
|
2386 |
} |
2387 |
|
2388 |
int bdrv_set_key(BlockDriverState *bs, const char *key) |
2389 |
{ |
2390 |
int ret;
|
2391 |
if (bs->backing_hd && bs->backing_hd->encrypted) {
|
2392 |
ret = bdrv_set_key(bs->backing_hd, key); |
2393 |
if (ret < 0) |
2394 |
return ret;
|
2395 |
if (!bs->encrypted)
|
2396 |
return 0; |
2397 |
} |
2398 |
if (!bs->encrypted) {
|
2399 |
return -EINVAL;
|
2400 |
} else if (!bs->drv || !bs->drv->bdrv_set_key) { |
2401 |
return -ENOMEDIUM;
|
2402 |
} |
2403 |
ret = bs->drv->bdrv_set_key(bs, key); |
2404 |
if (ret < 0) { |
2405 |
bs->valid_key = 0;
|
2406 |
} else if (!bs->valid_key) { |
2407 |
bs->valid_key = 1;
|
2408 |
/* call the change callback now, we skipped it on open */
|
2409 |
bdrv_dev_change_media_cb(bs, true);
|
2410 |
} |
2411 |
return ret;
|
2412 |
} |
2413 |
|
2414 |
const char *bdrv_get_format_name(BlockDriverState *bs) |
2415 |
{ |
2416 |
return bs->drv ? bs->drv->format_name : NULL; |
2417 |
} |
2418 |
|
2419 |
void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
2420 |
void *opaque)
|
2421 |
{ |
2422 |
BlockDriver *drv; |
2423 |
|
2424 |
QLIST_FOREACH(drv, &bdrv_drivers, list) { |
2425 |
it(opaque, drv->format_name); |
2426 |
} |
2427 |
} |
2428 |
|
2429 |
BlockDriverState *bdrv_find(const char *name) |
2430 |
{ |
2431 |
BlockDriverState *bs; |
2432 |
|
2433 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2434 |
if (!strcmp(name, bs->device_name)) {
|
2435 |
return bs;
|
2436 |
} |
2437 |
} |
2438 |
return NULL; |
2439 |
} |
2440 |
|
2441 |
BlockDriverState *bdrv_next(BlockDriverState *bs) |
2442 |
{ |
2443 |
if (!bs) {
|
2444 |
return QTAILQ_FIRST(&bdrv_states);
|
2445 |
} |
2446 |
return QTAILQ_NEXT(bs, list);
|
2447 |
} |
2448 |
|
2449 |
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque) |
2450 |
{ |
2451 |
BlockDriverState *bs; |
2452 |
|
2453 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2454 |
it(opaque, bs); |
2455 |
} |
2456 |
} |
2457 |
|
2458 |
const char *bdrv_get_device_name(BlockDriverState *bs) |
2459 |
{ |
2460 |
return bs->device_name;
|
2461 |
} |
2462 |
|
2463 |
int bdrv_get_flags(BlockDriverState *bs)
|
2464 |
{ |
2465 |
return bs->open_flags;
|
2466 |
} |
2467 |
|
2468 |
void bdrv_flush_all(void) |
2469 |
{ |
2470 |
BlockDriverState *bs; |
2471 |
|
2472 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2473 |
bdrv_flush(bs); |
2474 |
} |
2475 |
} |
2476 |
|
2477 |
int bdrv_has_zero_init(BlockDriverState *bs)
|
2478 |
{ |
2479 |
assert(bs->drv); |
2480 |
|
2481 |
if (bs->drv->bdrv_has_zero_init) {
|
2482 |
return bs->drv->bdrv_has_zero_init(bs);
|
2483 |
} |
2484 |
|
2485 |
return 1; |
2486 |
} |
2487 |
|
2488 |
typedef struct BdrvCoIsAllocatedData { |
2489 |
BlockDriverState *bs; |
2490 |
int64_t sector_num; |
2491 |
int nb_sectors;
|
2492 |
int *pnum;
|
2493 |
int ret;
|
2494 |
bool done;
|
2495 |
} BdrvCoIsAllocatedData; |
2496 |
|
2497 |
/*
|
2498 |
* Returns true iff the specified sector is present in the disk image. Drivers
|
2499 |
* not implementing the functionality are assumed to not support backing files,
|
2500 |
* hence all their sectors are reported as allocated.
|
2501 |
*
|
2502 |
* If 'sector_num' is beyond the end of the disk image the return value is 0
|
2503 |
* and 'pnum' is set to 0.
|
2504 |
*
|
2505 |
* 'pnum' is set to the number of sectors (including and immediately following
|
2506 |
* the specified sector) that are known to be in the same
|
2507 |
* allocated/unallocated state.
|
2508 |
*
|
2509 |
* 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
|
2510 |
* beyond the end of the disk image it will be clamped.
|
2511 |
*/
|
2512 |
int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
|
2513 |
int nb_sectors, int *pnum) |
2514 |
{ |
2515 |
int64_t n; |
2516 |
|
2517 |
if (sector_num >= bs->total_sectors) {
|
2518 |
*pnum = 0;
|
2519 |
return 0; |
2520 |
} |
2521 |
|
2522 |
n = bs->total_sectors - sector_num; |
2523 |
if (n < nb_sectors) {
|
2524 |
nb_sectors = n; |
2525 |
} |
2526 |
|
2527 |
if (!bs->drv->bdrv_co_is_allocated) {
|
2528 |
*pnum = nb_sectors; |
2529 |
return 1; |
2530 |
} |
2531 |
|
2532 |
return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
|
2533 |
} |
2534 |
|
2535 |
/* Coroutine wrapper for bdrv_is_allocated() */
|
2536 |
static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque) |
2537 |
{ |
2538 |
BdrvCoIsAllocatedData *data = opaque; |
2539 |
BlockDriverState *bs = data->bs; |
2540 |
|
2541 |
data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors, |
2542 |
data->pnum); |
2543 |
data->done = true;
|
2544 |
} |
2545 |
|
2546 |
/*
|
2547 |
* Synchronous wrapper around bdrv_co_is_allocated().
|
2548 |
*
|
2549 |
* See bdrv_co_is_allocated() for details.
|
2550 |
*/
|
2551 |
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors, |
2552 |
int *pnum)
|
2553 |
{ |
2554 |
Coroutine *co; |
2555 |
BdrvCoIsAllocatedData data = { |
2556 |
.bs = bs, |
2557 |
.sector_num = sector_num, |
2558 |
.nb_sectors = nb_sectors, |
2559 |
.pnum = pnum, |
2560 |
.done = false,
|
2561 |
}; |
2562 |
|
2563 |
co = qemu_coroutine_create(bdrv_is_allocated_co_entry); |
2564 |
qemu_coroutine_enter(co, &data); |
2565 |
while (!data.done) {
|
2566 |
qemu_aio_wait(); |
2567 |
} |
2568 |
return data.ret;
|
2569 |
} |
2570 |
|
2571 |
/*
|
2572 |
* Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
|
2573 |
*
|
2574 |
* Return true if the given sector is allocated in any image between
|
2575 |
* BASE and TOP (inclusive). BASE can be NULL to check if the given
|
2576 |
* sector is allocated in any image of the chain. Return false otherwise.
|
2577 |
*
|
2578 |
* 'pnum' is set to the number of sectors (including and immediately following
|
2579 |
* the specified sector) that are known to be in the same
|
2580 |
* allocated/unallocated state.
|
2581 |
*
|
2582 |
*/
|
2583 |
int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top,
|
2584 |
BlockDriverState *base, |
2585 |
int64_t sector_num, |
2586 |
int nb_sectors, int *pnum) |
2587 |
{ |
2588 |
BlockDriverState *intermediate; |
2589 |
int ret, n = nb_sectors;
|
2590 |
|
2591 |
intermediate = top; |
2592 |
while (intermediate && intermediate != base) {
|
2593 |
int pnum_inter;
|
2594 |
ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors, |
2595 |
&pnum_inter); |
2596 |
if (ret < 0) { |
2597 |
return ret;
|
2598 |
} else if (ret) { |
2599 |
*pnum = pnum_inter; |
2600 |
return 1; |
2601 |
} |
2602 |
|
2603 |
/*
|
2604 |
* [sector_num, nb_sectors] is unallocated on top but intermediate
|
2605 |
* might have
|
2606 |
*
|
2607 |
* [sector_num+x, nr_sectors] allocated.
|
2608 |
*/
|
2609 |
if (n > pnum_inter) {
|
2610 |
n = pnum_inter; |
2611 |
} |
2612 |
|
2613 |
intermediate = intermediate->backing_hd; |
2614 |
} |
2615 |
|
2616 |
*pnum = n; |
2617 |
return 0; |
2618 |
} |
2619 |
|
2620 |
BlockInfoList *qmp_query_block(Error **errp) |
2621 |
{ |
2622 |
BlockInfoList *head = NULL, *cur_item = NULL; |
2623 |
BlockDriverState *bs; |
2624 |
|
2625 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2626 |
BlockInfoList *info = g_malloc0(sizeof(*info));
|
2627 |
|
2628 |
info->value = g_malloc0(sizeof(*info->value));
|
2629 |
info->value->device = g_strdup(bs->device_name); |
2630 |
info->value->type = g_strdup("unknown");
|
2631 |
info->value->locked = bdrv_dev_is_medium_locked(bs); |
2632 |
info->value->removable = bdrv_dev_has_removable_media(bs); |
2633 |
|
2634 |
if (bdrv_dev_has_removable_media(bs)) {
|
2635 |
info->value->has_tray_open = true;
|
2636 |
info->value->tray_open = bdrv_dev_is_tray_open(bs); |
2637 |
} |
2638 |
|
2639 |
if (bdrv_iostatus_is_enabled(bs)) {
|
2640 |
info->value->has_io_status = true;
|
2641 |
info->value->io_status = bs->iostatus; |
2642 |
} |
2643 |
|
2644 |
if (bs->drv) {
|
2645 |
info->value->has_inserted = true;
|
2646 |
info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
|
2647 |
info->value->inserted->file = g_strdup(bs->filename); |
2648 |
info->value->inserted->ro = bs->read_only; |
2649 |
info->value->inserted->drv = g_strdup(bs->drv->format_name); |
2650 |
info->value->inserted->encrypted = bs->encrypted; |
2651 |
info->value->inserted->encryption_key_missing = bdrv_key_required(bs); |
2652 |
if (bs->backing_file[0]) { |
2653 |
info->value->inserted->has_backing_file = true;
|
2654 |
info->value->inserted->backing_file = g_strdup(bs->backing_file); |
2655 |
} |
2656 |
|
2657 |
info->value->inserted->backing_file_depth = |
2658 |
bdrv_get_backing_file_depth(bs); |
2659 |
|
2660 |
if (bs->io_limits_enabled) {
|
2661 |
info->value->inserted->bps = |
2662 |
bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; |
2663 |
info->value->inserted->bps_rd = |
2664 |
bs->io_limits.bps[BLOCK_IO_LIMIT_READ]; |
2665 |
info->value->inserted->bps_wr = |
2666 |
bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE]; |
2667 |
info->value->inserted->iops = |
2668 |
bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; |
2669 |
info->value->inserted->iops_rd = |
2670 |
bs->io_limits.iops[BLOCK_IO_LIMIT_READ]; |
2671 |
info->value->inserted->iops_wr = |
2672 |
bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE]; |
2673 |
} |
2674 |
} |
2675 |
|
2676 |
/* XXX: waiting for the qapi to support GSList */
|
2677 |
if (!cur_item) {
|
2678 |
head = cur_item = info; |
2679 |
} else {
|
2680 |
cur_item->next = info; |
2681 |
cur_item = info; |
2682 |
} |
2683 |
} |
2684 |
|
2685 |
return head;
|
2686 |
} |
2687 |
|
2688 |
/* Consider exposing this as a full fledged QMP command */
|
2689 |
static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp) |
2690 |
{ |
2691 |
BlockStats *s; |
2692 |
|
2693 |
s = g_malloc0(sizeof(*s));
|
2694 |
|
2695 |
if (bs->device_name[0]) { |
2696 |
s->has_device = true;
|
2697 |
s->device = g_strdup(bs->device_name); |
2698 |
} |
2699 |
|
2700 |
s->stats = g_malloc0(sizeof(*s->stats));
|
2701 |
s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ]; |
2702 |
s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE]; |
2703 |
s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ]; |
2704 |
s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE]; |
2705 |
s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE; |
2706 |
s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH]; |
2707 |
s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE]; |
2708 |
s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ]; |
2709 |
s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH]; |
2710 |
|
2711 |
if (bs->file) {
|
2712 |
s->has_parent = true;
|
2713 |
s->parent = qmp_query_blockstat(bs->file, NULL);
|
2714 |
} |
2715 |
|
2716 |
return s;
|
2717 |
} |
2718 |
|
2719 |
BlockStatsList *qmp_query_blockstats(Error **errp) |
2720 |
{ |
2721 |
BlockStatsList *head = NULL, *cur_item = NULL; |
2722 |
BlockDriverState *bs; |
2723 |
|
2724 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2725 |
BlockStatsList *info = g_malloc0(sizeof(*info));
|
2726 |
info->value = qmp_query_blockstat(bs, NULL);
|
2727 |
|
2728 |
/* XXX: waiting for the qapi to support GSList */
|
2729 |
if (!cur_item) {
|
2730 |
head = cur_item = info; |
2731 |
} else {
|
2732 |
cur_item->next = info; |
2733 |
cur_item = info; |
2734 |
} |
2735 |
} |
2736 |
|
2737 |
return head;
|
2738 |
} |
2739 |
|
2740 |
const char *bdrv_get_encrypted_filename(BlockDriverState *bs) |
2741 |
{ |
2742 |
if (bs->backing_hd && bs->backing_hd->encrypted)
|
2743 |
return bs->backing_file;
|
2744 |
else if (bs->encrypted) |
2745 |
return bs->filename;
|
2746 |
else
|
2747 |
return NULL; |
2748 |
} |
2749 |
|
2750 |
void bdrv_get_backing_filename(BlockDriverState *bs,
|
2751 |
char *filename, int filename_size) |
2752 |
{ |
2753 |
pstrcpy(filename, filename_size, bs->backing_file); |
2754 |
} |
2755 |
|
2756 |
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
|
2757 |
const uint8_t *buf, int nb_sectors) |
2758 |
{ |
2759 |
BlockDriver *drv = bs->drv; |
2760 |
if (!drv)
|
2761 |
return -ENOMEDIUM;
|
2762 |
if (!drv->bdrv_write_compressed)
|
2763 |
return -ENOTSUP;
|
2764 |
if (bdrv_check_request(bs, sector_num, nb_sectors))
|
2765 |
return -EIO;
|
2766 |
|
2767 |
if (bs->dirty_bitmap) {
|
2768 |
set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
|
2769 |
} |
2770 |
|
2771 |
return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
|
2772 |
} |
2773 |
|
2774 |
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
|
2775 |
{ |
2776 |
BlockDriver *drv = bs->drv; |
2777 |
if (!drv)
|
2778 |
return -ENOMEDIUM;
|
2779 |
if (!drv->bdrv_get_info)
|
2780 |
return -ENOTSUP;
|
2781 |
memset(bdi, 0, sizeof(*bdi)); |
2782 |
return drv->bdrv_get_info(bs, bdi);
|
2783 |
} |
2784 |
|
2785 |
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2786 |
int64_t pos, int size)
|
2787 |
{ |
2788 |
BlockDriver *drv = bs->drv; |
2789 |
if (!drv)
|
2790 |
return -ENOMEDIUM;
|
2791 |
if (drv->bdrv_save_vmstate)
|
2792 |
return drv->bdrv_save_vmstate(bs, buf, pos, size);
|
2793 |
if (bs->file)
|
2794 |
return bdrv_save_vmstate(bs->file, buf, pos, size);
|
2795 |
return -ENOTSUP;
|
2796 |
} |
2797 |
|
2798 |
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
|
2799 |
int64_t pos, int size)
|
2800 |
{ |
2801 |
BlockDriver *drv = bs->drv; |
2802 |
if (!drv)
|
2803 |
return -ENOMEDIUM;
|
2804 |
if (drv->bdrv_load_vmstate)
|
2805 |
return drv->bdrv_load_vmstate(bs, buf, pos, size);
|
2806 |
if (bs->file)
|
2807 |
return bdrv_load_vmstate(bs->file, buf, pos, size);
|
2808 |
return -ENOTSUP;
|
2809 |
} |
2810 |
|
2811 |
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
|
2812 |
{ |
2813 |
BlockDriver *drv = bs->drv; |
2814 |
|
2815 |
if (!drv || !drv->bdrv_debug_event) {
|
2816 |
return;
|
2817 |
} |
2818 |
|
2819 |
drv->bdrv_debug_event(bs, event); |
2820 |
|
2821 |
} |
2822 |
|
2823 |
/**************************************************************/
|
2824 |
/* handling of snapshots */
|
2825 |
|
2826 |
int bdrv_can_snapshot(BlockDriverState *bs)
|
2827 |
{ |
2828 |
BlockDriver *drv = bs->drv; |
2829 |
if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
|
2830 |
return 0; |
2831 |
} |
2832 |
|
2833 |
if (!drv->bdrv_snapshot_create) {
|
2834 |
if (bs->file != NULL) { |
2835 |
return bdrv_can_snapshot(bs->file);
|
2836 |
} |
2837 |
return 0; |
2838 |
} |
2839 |
|
2840 |
return 1; |
2841 |
} |
2842 |
|
2843 |
int bdrv_is_snapshot(BlockDriverState *bs)
|
2844 |
{ |
2845 |
return !!(bs->open_flags & BDRV_O_SNAPSHOT);
|
2846 |
} |
2847 |
|
2848 |
BlockDriverState *bdrv_snapshots(void)
|
2849 |
{ |
2850 |
BlockDriverState *bs; |
2851 |
|
2852 |
if (bs_snapshots) {
|
2853 |
return bs_snapshots;
|
2854 |
} |
2855 |
|
2856 |
bs = NULL;
|
2857 |
while ((bs = bdrv_next(bs))) {
|
2858 |
if (bdrv_can_snapshot(bs)) {
|
2859 |
bs_snapshots = bs; |
2860 |
return bs;
|
2861 |
} |
2862 |
} |
2863 |
return NULL; |
2864 |
} |
2865 |
|
2866 |
int bdrv_snapshot_create(BlockDriverState *bs,
|
2867 |
QEMUSnapshotInfo *sn_info) |
2868 |
{ |
2869 |
BlockDriver *drv = bs->drv; |
2870 |
if (!drv)
|
2871 |
return -ENOMEDIUM;
|
2872 |
if (drv->bdrv_snapshot_create)
|
2873 |
return drv->bdrv_snapshot_create(bs, sn_info);
|
2874 |
if (bs->file)
|
2875 |
return bdrv_snapshot_create(bs->file, sn_info);
|
2876 |
return -ENOTSUP;
|
2877 |
} |
2878 |
|
2879 |
int bdrv_snapshot_goto(BlockDriverState *bs,
|
2880 |
const char *snapshot_id) |
2881 |
{ |
2882 |
BlockDriver *drv = bs->drv; |
2883 |
int ret, open_ret;
|
2884 |
|
2885 |
if (!drv)
|
2886 |
return -ENOMEDIUM;
|
2887 |
if (drv->bdrv_snapshot_goto)
|
2888 |
return drv->bdrv_snapshot_goto(bs, snapshot_id);
|
2889 |
|
2890 |
if (bs->file) {
|
2891 |
drv->bdrv_close(bs); |
2892 |
ret = bdrv_snapshot_goto(bs->file, snapshot_id); |
2893 |
open_ret = drv->bdrv_open(bs, bs->open_flags); |
2894 |
if (open_ret < 0) { |
2895 |
bdrv_delete(bs->file); |
2896 |
bs->drv = NULL;
|
2897 |
return open_ret;
|
2898 |
} |
2899 |
return ret;
|
2900 |
} |
2901 |
|
2902 |
return -ENOTSUP;
|
2903 |
} |
2904 |
|
2905 |
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) |
2906 |
{ |
2907 |
BlockDriver *drv = bs->drv; |
2908 |
if (!drv)
|
2909 |
return -ENOMEDIUM;
|
2910 |
if (drv->bdrv_snapshot_delete)
|
2911 |
return drv->bdrv_snapshot_delete(bs, snapshot_id);
|
2912 |
if (bs->file)
|
2913 |
return bdrv_snapshot_delete(bs->file, snapshot_id);
|
2914 |
return -ENOTSUP;
|
2915 |
} |
2916 |
|
2917 |
int bdrv_snapshot_list(BlockDriverState *bs,
|
2918 |
QEMUSnapshotInfo **psn_info) |
2919 |
{ |
2920 |
BlockDriver *drv = bs->drv; |
2921 |
if (!drv)
|
2922 |
return -ENOMEDIUM;
|
2923 |
if (drv->bdrv_snapshot_list)
|
2924 |
return drv->bdrv_snapshot_list(bs, psn_info);
|
2925 |
if (bs->file)
|
2926 |
return bdrv_snapshot_list(bs->file, psn_info);
|
2927 |
return -ENOTSUP;
|
2928 |
} |
2929 |
|
2930 |
int bdrv_snapshot_load_tmp(BlockDriverState *bs,
|
2931 |
const char *snapshot_name) |
2932 |
{ |
2933 |
BlockDriver *drv = bs->drv; |
2934 |
if (!drv) {
|
2935 |
return -ENOMEDIUM;
|
2936 |
} |
2937 |
if (!bs->read_only) {
|
2938 |
return -EINVAL;
|
2939 |
} |
2940 |
if (drv->bdrv_snapshot_load_tmp) {
|
2941 |
return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
|
2942 |
} |
2943 |
return -ENOTSUP;
|
2944 |
} |
2945 |
|
2946 |
BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, |
2947 |
const char *backing_file) |
2948 |
{ |
2949 |
if (!bs->drv) {
|
2950 |
return NULL; |
2951 |
} |
2952 |
|
2953 |
if (bs->backing_hd) {
|
2954 |
if (strcmp(bs->backing_file, backing_file) == 0) { |
2955 |
return bs->backing_hd;
|
2956 |
} else {
|
2957 |
return bdrv_find_backing_image(bs->backing_hd, backing_file);
|
2958 |
} |
2959 |
} |
2960 |
|
2961 |
return NULL; |
2962 |
} |
2963 |
|
2964 |
int bdrv_get_backing_file_depth(BlockDriverState *bs)
|
2965 |
{ |
2966 |
if (!bs->drv) {
|
2967 |
return 0; |
2968 |
} |
2969 |
|
2970 |
if (!bs->backing_hd) {
|
2971 |
return 0; |
2972 |
} |
2973 |
|
2974 |
return 1 + bdrv_get_backing_file_depth(bs->backing_hd); |
2975 |
} |
2976 |
|
2977 |
#define NB_SUFFIXES 4 |
2978 |
|
2979 |
char *get_human_readable_size(char *buf, int buf_size, int64_t size) |
2980 |
{ |
2981 |
static const char suffixes[NB_SUFFIXES] = "KMGT"; |
2982 |
int64_t base; |
2983 |
int i;
|
2984 |
|
2985 |
if (size <= 999) { |
2986 |
snprintf(buf, buf_size, "%" PRId64, size);
|
2987 |
} else {
|
2988 |
base = 1024;
|
2989 |
for(i = 0; i < NB_SUFFIXES; i++) { |
2990 |
if (size < (10 * base)) { |
2991 |
snprintf(buf, buf_size, "%0.1f%c",
|
2992 |
(double)size / base,
|
2993 |
suffixes[i]); |
2994 |
break;
|
2995 |
} else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) { |
2996 |
snprintf(buf, buf_size, "%" PRId64 "%c", |
2997 |
((size + (base >> 1)) / base),
|
2998 |
suffixes[i]); |
2999 |
break;
|
3000 |
} |
3001 |
base = base * 1024;
|
3002 |
} |
3003 |
} |
3004 |
return buf;
|
3005 |
} |
3006 |
|
3007 |
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn) |
3008 |
{ |
3009 |
char buf1[128], date_buf[128], clock_buf[128]; |
3010 |
#ifdef _WIN32
|
3011 |
struct tm *ptm;
|
3012 |
#else
|
3013 |
struct tm tm;
|
3014 |
#endif
|
3015 |
time_t ti; |
3016 |
int64_t secs; |
3017 |
|
3018 |
if (!sn) {
|
3019 |
snprintf(buf, buf_size, |
3020 |
"%-10s%-20s%7s%20s%15s",
|
3021 |
"ID", "TAG", "VM SIZE", "DATE", "VM CLOCK"); |
3022 |
} else {
|
3023 |
ti = sn->date_sec; |
3024 |
#ifdef _WIN32
|
3025 |
ptm = localtime(&ti); |
3026 |
strftime(date_buf, sizeof(date_buf),
|
3027 |
"%Y-%m-%d %H:%M:%S", ptm);
|
3028 |
#else
|
3029 |
localtime_r(&ti, &tm); |
3030 |
strftime(date_buf, sizeof(date_buf),
|
3031 |
"%Y-%m-%d %H:%M:%S", &tm);
|
3032 |
#endif
|
3033 |
secs = sn->vm_clock_nsec / 1000000000;
|
3034 |
snprintf(clock_buf, sizeof(clock_buf),
|
3035 |
"%02d:%02d:%02d.%03d",
|
3036 |
(int)(secs / 3600), |
3037 |
(int)((secs / 60) % 60), |
3038 |
(int)(secs % 60), |
3039 |
(int)((sn->vm_clock_nsec / 1000000) % 1000)); |
3040 |
snprintf(buf, buf_size, |
3041 |
"%-10s%-20s%7s%20s%15s",
|
3042 |
sn->id_str, sn->name, |
3043 |
get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
|
3044 |
date_buf, |
3045 |
clock_buf); |
3046 |
} |
3047 |
return buf;
|
3048 |
} |
3049 |
|
3050 |
/**************************************************************/
|
3051 |
/* async I/Os */
|
3052 |
|
3053 |
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, |
3054 |
QEMUIOVector *qiov, int nb_sectors,
|
3055 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3056 |
{ |
3057 |
trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); |
3058 |
|
3059 |
return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
|
3060 |
cb, opaque, false);
|
3061 |
} |
3062 |
|
3063 |
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, |
3064 |
QEMUIOVector *qiov, int nb_sectors,
|
3065 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3066 |
{ |
3067 |
trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); |
3068 |
|
3069 |
return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
|
3070 |
cb, opaque, true);
|
3071 |
} |
3072 |
|
3073 |
|
3074 |
typedef struct MultiwriteCB { |
3075 |
int error;
|
3076 |
int num_requests;
|
3077 |
int num_callbacks;
|
3078 |
struct {
|
3079 |
BlockDriverCompletionFunc *cb; |
3080 |
void *opaque;
|
3081 |
QEMUIOVector *free_qiov; |
3082 |
} callbacks[]; |
3083 |
} MultiwriteCB; |
3084 |
|
3085 |
static void multiwrite_user_cb(MultiwriteCB *mcb) |
3086 |
{ |
3087 |
int i;
|
3088 |
|
3089 |
for (i = 0; i < mcb->num_callbacks; i++) { |
3090 |
mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); |
3091 |
if (mcb->callbacks[i].free_qiov) {
|
3092 |
qemu_iovec_destroy(mcb->callbacks[i].free_qiov); |
3093 |
} |
3094 |
g_free(mcb->callbacks[i].free_qiov); |
3095 |
} |
3096 |
} |
3097 |
|
3098 |
static void multiwrite_cb(void *opaque, int ret) |
3099 |
{ |
3100 |
MultiwriteCB *mcb = opaque; |
3101 |
|
3102 |
trace_multiwrite_cb(mcb, ret); |
3103 |
|
3104 |
if (ret < 0 && !mcb->error) { |
3105 |
mcb->error = ret; |
3106 |
} |
3107 |
|
3108 |
mcb->num_requests--; |
3109 |
if (mcb->num_requests == 0) { |
3110 |
multiwrite_user_cb(mcb); |
3111 |
g_free(mcb); |
3112 |
} |
3113 |
} |
3114 |
|
3115 |
static int multiwrite_req_compare(const void *a, const void *b) |
3116 |
{ |
3117 |
const BlockRequest *req1 = a, *req2 = b;
|
3118 |
|
3119 |
/*
|
3120 |
* Note that we can't simply subtract req2->sector from req1->sector
|
3121 |
* here as that could overflow the return value.
|
3122 |
*/
|
3123 |
if (req1->sector > req2->sector) {
|
3124 |
return 1; |
3125 |
} else if (req1->sector < req2->sector) { |
3126 |
return -1; |
3127 |
} else {
|
3128 |
return 0; |
3129 |
} |
3130 |
} |
3131 |
|
3132 |
/*
|
3133 |
* Takes a bunch of requests and tries to merge them. Returns the number of
|
3134 |
* requests that remain after merging.
|
3135 |
*/
|
3136 |
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, |
3137 |
int num_reqs, MultiwriteCB *mcb)
|
3138 |
{ |
3139 |
int i, outidx;
|
3140 |
|
3141 |
// Sort requests by start sector
|
3142 |
qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
|
3143 |
|
3144 |
// Check if adjacent requests touch the same clusters. If so, combine them,
|
3145 |
// filling up gaps with zero sectors.
|
3146 |
outidx = 0;
|
3147 |
for (i = 1; i < num_reqs; i++) { |
3148 |
int merge = 0; |
3149 |
int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; |
3150 |
|
3151 |
// Handle exactly sequential writes and overlapping writes.
|
3152 |
if (reqs[i].sector <= oldreq_last) {
|
3153 |
merge = 1;
|
3154 |
} |
3155 |
|
3156 |
if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { |
3157 |
merge = 0;
|
3158 |
} |
3159 |
|
3160 |
if (merge) {
|
3161 |
size_t size; |
3162 |
QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
|
3163 |
qemu_iovec_init(qiov, |
3164 |
reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
|
3165 |
|
3166 |
// Add the first request to the merged one. If the requests are
|
3167 |
// overlapping, drop the last sectors of the first request.
|
3168 |
size = (reqs[i].sector - reqs[outidx].sector) << 9;
|
3169 |
qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
|
3170 |
|
3171 |
// We should need to add any zeros between the two requests
|
3172 |
assert (reqs[i].sector <= oldreq_last); |
3173 |
|
3174 |
// Add the second request
|
3175 |
qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
|
3176 |
|
3177 |
reqs[outidx].nb_sectors = qiov->size >> 9;
|
3178 |
reqs[outidx].qiov = qiov; |
3179 |
|
3180 |
mcb->callbacks[i].free_qiov = reqs[outidx].qiov; |
3181 |
} else {
|
3182 |
outidx++; |
3183 |
reqs[outidx].sector = reqs[i].sector; |
3184 |
reqs[outidx].nb_sectors = reqs[i].nb_sectors; |
3185 |
reqs[outidx].qiov = reqs[i].qiov; |
3186 |
} |
3187 |
} |
3188 |
|
3189 |
return outidx + 1; |
3190 |
} |
3191 |
|
3192 |
/*
|
3193 |
* Submit multiple AIO write requests at once.
|
3194 |
*
|
3195 |
* On success, the function returns 0 and all requests in the reqs array have
|
3196 |
* been submitted. In error case this function returns -1, and any of the
|
3197 |
* requests may or may not be submitted yet. In particular, this means that the
|
3198 |
* callback will be called for some of the requests, for others it won't. The
|
3199 |
* caller must check the error field of the BlockRequest to wait for the right
|
3200 |
* callbacks (if error != 0, no callback will be called).
|
3201 |
*
|
3202 |
* The implementation may modify the contents of the reqs array, e.g. to merge
|
3203 |
* requests. However, the fields opaque and error are left unmodified as they
|
3204 |
* are used to signal failure for a single request to the caller.
|
3205 |
*/
|
3206 |
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) |
3207 |
{ |
3208 |
MultiwriteCB *mcb; |
3209 |
int i;
|
3210 |
|
3211 |
/* don't submit writes if we don't have a medium */
|
3212 |
if (bs->drv == NULL) { |
3213 |
for (i = 0; i < num_reqs; i++) { |
3214 |
reqs[i].error = -ENOMEDIUM; |
3215 |
} |
3216 |
return -1; |
3217 |
} |
3218 |
|
3219 |
if (num_reqs == 0) { |
3220 |
return 0; |
3221 |
} |
3222 |
|
3223 |
// Create MultiwriteCB structure
|
3224 |
mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); |
3225 |
mcb->num_requests = 0;
|
3226 |
mcb->num_callbacks = num_reqs; |
3227 |
|
3228 |
for (i = 0; i < num_reqs; i++) { |
3229 |
mcb->callbacks[i].cb = reqs[i].cb; |
3230 |
mcb->callbacks[i].opaque = reqs[i].opaque; |
3231 |
} |
3232 |
|
3233 |
// Check for mergable requests
|
3234 |
num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); |
3235 |
|
3236 |
trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); |
3237 |
|
3238 |
/* Run the aio requests. */
|
3239 |
mcb->num_requests = num_reqs; |
3240 |
for (i = 0; i < num_reqs; i++) { |
3241 |
bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov, |
3242 |
reqs[i].nb_sectors, multiwrite_cb, mcb); |
3243 |
} |
3244 |
|
3245 |
return 0; |
3246 |
} |
3247 |
|
3248 |
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
|
3249 |
{ |
3250 |
acb->pool->cancel(acb); |
3251 |
} |
3252 |
|
3253 |
/* block I/O throttling */
|
3254 |
static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors, |
3255 |
bool is_write, double elapsed_time, uint64_t *wait) |
3256 |
{ |
3257 |
uint64_t bps_limit = 0;
|
3258 |
double bytes_limit, bytes_base, bytes_res;
|
3259 |
double slice_time, wait_time;
|
3260 |
|
3261 |
if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
|
3262 |
bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; |
3263 |
} else if (bs->io_limits.bps[is_write]) { |
3264 |
bps_limit = bs->io_limits.bps[is_write]; |
3265 |
} else {
|
3266 |
if (wait) {
|
3267 |
*wait = 0;
|
3268 |
} |
3269 |
|
3270 |
return false; |
3271 |
} |
3272 |
|
3273 |
slice_time = bs->slice_end - bs->slice_start; |
3274 |
slice_time /= (NANOSECONDS_PER_SECOND); |
3275 |
bytes_limit = bps_limit * slice_time; |
3276 |
bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write]; |
3277 |
if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
|
3278 |
bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write]; |
3279 |
} |
3280 |
|
3281 |
/* bytes_base: the bytes of data which have been read/written; and
|
3282 |
* it is obtained from the history statistic info.
|
3283 |
* bytes_res: the remaining bytes of data which need to be read/written.
|
3284 |
* (bytes_base + bytes_res) / bps_limit: used to calcuate
|
3285 |
* the total time for completing reading/writting all data.
|
3286 |
*/
|
3287 |
bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
|
3288 |
|
3289 |
if (bytes_base + bytes_res <= bytes_limit) {
|
3290 |
if (wait) {
|
3291 |
*wait = 0;
|
3292 |
} |
3293 |
|
3294 |
return false; |
3295 |
} |
3296 |
|
3297 |
/* Calc approx time to dispatch */
|
3298 |
wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time; |
3299 |
|
3300 |
/* When the I/O rate at runtime exceeds the limits,
|
3301 |
* bs->slice_end need to be extended in order that the current statistic
|
3302 |
* info can be kept until the timer fire, so it is increased and tuned
|
3303 |
* based on the result of experiment.
|
3304 |
*/
|
3305 |
bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3306 |
bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
|
3307 |
if (wait) {
|
3308 |
*wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3309 |
} |
3310 |
|
3311 |
return true; |
3312 |
} |
3313 |
|
3314 |
static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write, |
3315 |
double elapsed_time, uint64_t *wait)
|
3316 |
{ |
3317 |
uint64_t iops_limit = 0;
|
3318 |
double ios_limit, ios_base;
|
3319 |
double slice_time, wait_time;
|
3320 |
|
3321 |
if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
|
3322 |
iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; |
3323 |
} else if (bs->io_limits.iops[is_write]) { |
3324 |
iops_limit = bs->io_limits.iops[is_write]; |
3325 |
} else {
|
3326 |
if (wait) {
|
3327 |
*wait = 0;
|
3328 |
} |
3329 |
|
3330 |
return false; |
3331 |
} |
3332 |
|
3333 |
slice_time = bs->slice_end - bs->slice_start; |
3334 |
slice_time /= (NANOSECONDS_PER_SECOND); |
3335 |
ios_limit = iops_limit * slice_time; |
3336 |
ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write]; |
3337 |
if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
|
3338 |
ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write]; |
3339 |
} |
3340 |
|
3341 |
if (ios_base + 1 <= ios_limit) { |
3342 |
if (wait) {
|
3343 |
*wait = 0;
|
3344 |
} |
3345 |
|
3346 |
return false; |
3347 |
} |
3348 |
|
3349 |
/* Calc approx time to dispatch */
|
3350 |
wait_time = (ios_base + 1) / iops_limit;
|
3351 |
if (wait_time > elapsed_time) {
|
3352 |
wait_time = wait_time - elapsed_time; |
3353 |
} else {
|
3354 |
wait_time = 0;
|
3355 |
} |
3356 |
|
3357 |
bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3358 |
bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
|
3359 |
if (wait) {
|
3360 |
*wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3361 |
} |
3362 |
|
3363 |
return true; |
3364 |
} |
3365 |
|
3366 |
static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors, |
3367 |
bool is_write, int64_t *wait)
|
3368 |
{ |
3369 |
int64_t now, max_wait; |
3370 |
uint64_t bps_wait = 0, iops_wait = 0; |
3371 |
double elapsed_time;
|
3372 |
int bps_ret, iops_ret;
|
3373 |
|
3374 |
now = qemu_get_clock_ns(vm_clock); |
3375 |
if ((bs->slice_start < now)
|
3376 |
&& (bs->slice_end > now)) { |
3377 |
bs->slice_end = now + bs->slice_time; |
3378 |
} else {
|
3379 |
bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
|
3380 |
bs->slice_start = now; |
3381 |
bs->slice_end = now + bs->slice_time; |
3382 |
|
3383 |
bs->io_base.bytes[is_write] = bs->nr_bytes[is_write]; |
3384 |
bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write]; |
3385 |
|
3386 |
bs->io_base.ios[is_write] = bs->nr_ops[is_write]; |
3387 |
bs->io_base.ios[!is_write] = bs->nr_ops[!is_write]; |
3388 |
} |
3389 |
|
3390 |
elapsed_time = now - bs->slice_start; |
3391 |
elapsed_time /= (NANOSECONDS_PER_SECOND); |
3392 |
|
3393 |
bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors, |
3394 |
is_write, elapsed_time, &bps_wait); |
3395 |
iops_ret = bdrv_exceed_iops_limits(bs, is_write, |
3396 |
elapsed_time, &iops_wait); |
3397 |
if (bps_ret || iops_ret) {
|
3398 |
max_wait = bps_wait > iops_wait ? bps_wait : iops_wait; |
3399 |
if (wait) {
|
3400 |
*wait = max_wait; |
3401 |
} |
3402 |
|
3403 |
now = qemu_get_clock_ns(vm_clock); |
3404 |
if (bs->slice_end < now + max_wait) {
|
3405 |
bs->slice_end = now + max_wait; |
3406 |
} |
3407 |
|
3408 |
return true; |
3409 |
} |
3410 |
|
3411 |
if (wait) {
|
3412 |
*wait = 0;
|
3413 |
} |
3414 |
|
3415 |
return false; |
3416 |
} |
3417 |
|
3418 |
/**************************************************************/
|
3419 |
/* async block device emulation */
|
3420 |
|
3421 |
typedef struct BlockDriverAIOCBSync { |
3422 |
BlockDriverAIOCB common; |
3423 |
QEMUBH *bh; |
3424 |
int ret;
|
3425 |
/* vector translation state */
|
3426 |
QEMUIOVector *qiov; |
3427 |
uint8_t *bounce; |
3428 |
int is_write;
|
3429 |
} BlockDriverAIOCBSync; |
3430 |
|
3431 |
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb) |
3432 |
{ |
3433 |
BlockDriverAIOCBSync *acb = |
3434 |
container_of(blockacb, BlockDriverAIOCBSync, common); |
3435 |
qemu_bh_delete(acb->bh); |
3436 |
acb->bh = NULL;
|
3437 |
qemu_aio_release(acb); |
3438 |
} |
3439 |
|
3440 |
static AIOPool bdrv_em_aio_pool = {
|
3441 |
.aiocb_size = sizeof(BlockDriverAIOCBSync),
|
3442 |
.cancel = bdrv_aio_cancel_em, |
3443 |
}; |
3444 |
|
3445 |
static void bdrv_aio_bh_cb(void *opaque) |
3446 |
{ |
3447 |
BlockDriverAIOCBSync *acb = opaque; |
3448 |
|
3449 |
if (!acb->is_write)
|
3450 |
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
|
3451 |
qemu_vfree(acb->bounce); |
3452 |
acb->common.cb(acb->common.opaque, acb->ret); |
3453 |
qemu_bh_delete(acb->bh); |
3454 |
acb->bh = NULL;
|
3455 |
qemu_aio_release(acb); |
3456 |
} |
3457 |
|
3458 |
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
|
3459 |
int64_t sector_num, |
3460 |
QEMUIOVector *qiov, |
3461 |
int nb_sectors,
|
3462 |
BlockDriverCompletionFunc *cb, |
3463 |
void *opaque,
|
3464 |
int is_write)
|
3465 |
|
3466 |
{ |
3467 |
BlockDriverAIOCBSync *acb; |
3468 |
|
3469 |
acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque); |
3470 |
acb->is_write = is_write; |
3471 |
acb->qiov = qiov; |
3472 |
acb->bounce = qemu_blockalign(bs, qiov->size); |
3473 |
acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb); |
3474 |
|
3475 |
if (is_write) {
|
3476 |
qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
|
3477 |
acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); |
3478 |
} else {
|
3479 |
acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); |
3480 |
} |
3481 |
|
3482 |
qemu_bh_schedule(acb->bh); |
3483 |
|
3484 |
return &acb->common;
|
3485 |
} |
3486 |
|
3487 |
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
|
3488 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
3489 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3490 |
{ |
3491 |
return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); |
3492 |
} |
3493 |
|
3494 |
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
|
3495 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
3496 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3497 |
{ |
3498 |
return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); |
3499 |
} |
3500 |
|
3501 |
|
3502 |
typedef struct BlockDriverAIOCBCoroutine { |
3503 |
BlockDriverAIOCB common; |
3504 |
BlockRequest req; |
3505 |
bool is_write;
|
3506 |
QEMUBH* bh; |
3507 |
} BlockDriverAIOCBCoroutine; |
3508 |
|
3509 |
static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb) |
3510 |
{ |
3511 |
qemu_aio_flush(); |
3512 |
} |
3513 |
|
3514 |
static AIOPool bdrv_em_co_aio_pool = {
|
3515 |
.aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
|
3516 |
.cancel = bdrv_aio_co_cancel_em, |
3517 |
}; |
3518 |
|
3519 |
static void bdrv_co_em_bh(void *opaque) |
3520 |
{ |
3521 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3522 |
|
3523 |
acb->common.cb(acb->common.opaque, acb->req.error); |
3524 |
qemu_bh_delete(acb->bh); |
3525 |
qemu_aio_release(acb); |
3526 |
} |
3527 |
|
3528 |
/* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
|
3529 |
static void coroutine_fn bdrv_co_do_rw(void *opaque) |
3530 |
{ |
3531 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3532 |
BlockDriverState *bs = acb->common.bs; |
3533 |
|
3534 |
if (!acb->is_write) {
|
3535 |
acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, |
3536 |
acb->req.nb_sectors, acb->req.qiov, 0);
|
3537 |
} else {
|
3538 |
acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, |
3539 |
acb->req.nb_sectors, acb->req.qiov, 0);
|
3540 |
} |
3541 |
|
3542 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3543 |
qemu_bh_schedule(acb->bh); |
3544 |
} |
3545 |
|
3546 |
static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
|
3547 |
int64_t sector_num, |
3548 |
QEMUIOVector *qiov, |
3549 |
int nb_sectors,
|
3550 |
BlockDriverCompletionFunc *cb, |
3551 |
void *opaque,
|
3552 |
bool is_write)
|
3553 |
{ |
3554 |
Coroutine *co; |
3555 |
BlockDriverAIOCBCoroutine *acb; |
3556 |
|
3557 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3558 |
acb->req.sector = sector_num; |
3559 |
acb->req.nb_sectors = nb_sectors; |
3560 |
acb->req.qiov = qiov; |
3561 |
acb->is_write = is_write; |
3562 |
|
3563 |
co = qemu_coroutine_create(bdrv_co_do_rw); |
3564 |
qemu_coroutine_enter(co, acb); |
3565 |
|
3566 |
return &acb->common;
|
3567 |
} |
3568 |
|
3569 |
static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) |
3570 |
{ |
3571 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3572 |
BlockDriverState *bs = acb->common.bs; |
3573 |
|
3574 |
acb->req.error = bdrv_co_flush(bs); |
3575 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3576 |
qemu_bh_schedule(acb->bh); |
3577 |
} |
3578 |
|
3579 |
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs, |
3580 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3581 |
{ |
3582 |
trace_bdrv_aio_flush(bs, opaque); |
3583 |
|
3584 |
Coroutine *co; |
3585 |
BlockDriverAIOCBCoroutine *acb; |
3586 |
|
3587 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3588 |
co = qemu_coroutine_create(bdrv_aio_flush_co_entry); |
3589 |
qemu_coroutine_enter(co, acb); |
3590 |
|
3591 |
return &acb->common;
|
3592 |
} |
3593 |
|
3594 |
static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) |
3595 |
{ |
3596 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3597 |
BlockDriverState *bs = acb->common.bs; |
3598 |
|
3599 |
acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); |
3600 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3601 |
qemu_bh_schedule(acb->bh); |
3602 |
} |
3603 |
|
3604 |
BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs, |
3605 |
int64_t sector_num, int nb_sectors,
|
3606 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3607 |
{ |
3608 |
Coroutine *co; |
3609 |
BlockDriverAIOCBCoroutine *acb; |
3610 |
|
3611 |
trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); |
3612 |
|
3613 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3614 |
acb->req.sector = sector_num; |
3615 |
acb->req.nb_sectors = nb_sectors; |
3616 |
co = qemu_coroutine_create(bdrv_aio_discard_co_entry); |
3617 |
qemu_coroutine_enter(co, acb); |
3618 |
|
3619 |
return &acb->common;
|
3620 |
} |
3621 |
|
3622 |
void bdrv_init(void) |
3623 |
{ |
3624 |
module_call_init(MODULE_INIT_BLOCK); |
3625 |
} |
3626 |
|
3627 |
void bdrv_init_with_whitelist(void) |
3628 |
{ |
3629 |
use_bdrv_whitelist = 1;
|
3630 |
bdrv_init(); |
3631 |
} |
3632 |
|
3633 |
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
|
3634 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3635 |
{ |
3636 |
BlockDriverAIOCB *acb; |
3637 |
|
3638 |
if (pool->free_aiocb) {
|
3639 |
acb = pool->free_aiocb; |
3640 |
pool->free_aiocb = acb->next; |
3641 |
} else {
|
3642 |
acb = g_malloc0(pool->aiocb_size); |
3643 |
acb->pool = pool; |
3644 |
} |
3645 |
acb->bs = bs; |
3646 |
acb->cb = cb; |
3647 |
acb->opaque = opaque; |
3648 |
return acb;
|
3649 |
} |
3650 |
|
3651 |
void qemu_aio_release(void *p) |
3652 |
{ |
3653 |
BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p; |
3654 |
AIOPool *pool = acb->pool; |
3655 |
acb->next = pool->free_aiocb; |
3656 |
pool->free_aiocb = acb; |
3657 |
} |
3658 |
|
3659 |
/**************************************************************/
|
3660 |
/* Coroutine block device emulation */
|
3661 |
|
3662 |
typedef struct CoroutineIOCompletion { |
3663 |
Coroutine *coroutine; |
3664 |
int ret;
|
3665 |
} CoroutineIOCompletion; |
3666 |
|
3667 |
static void bdrv_co_io_em_complete(void *opaque, int ret) |
3668 |
{ |
3669 |
CoroutineIOCompletion *co = opaque; |
3670 |
|
3671 |
co->ret = ret; |
3672 |
qemu_coroutine_enter(co->coroutine, NULL);
|
3673 |
} |
3674 |
|
3675 |
static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, |
3676 |
int nb_sectors, QEMUIOVector *iov,
|
3677 |
bool is_write)
|
3678 |
{ |
3679 |
CoroutineIOCompletion co = { |
3680 |
.coroutine = qemu_coroutine_self(), |
3681 |
}; |
3682 |
BlockDriverAIOCB *acb; |
3683 |
|
3684 |
if (is_write) {
|
3685 |
acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, |
3686 |
bdrv_co_io_em_complete, &co); |
3687 |
} else {
|
3688 |
acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, |
3689 |
bdrv_co_io_em_complete, &co); |
3690 |
} |
3691 |
|
3692 |
trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); |
3693 |
if (!acb) {
|
3694 |
return -EIO;
|
3695 |
} |
3696 |
qemu_coroutine_yield(); |
3697 |
|
3698 |
return co.ret;
|
3699 |
} |
3700 |
|
3701 |
static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, |
3702 |
int64_t sector_num, int nb_sectors,
|
3703 |
QEMUIOVector *iov) |
3704 |
{ |
3705 |
return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); |
3706 |
} |
3707 |
|
3708 |
static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, |
3709 |
int64_t sector_num, int nb_sectors,
|
3710 |
QEMUIOVector *iov) |
3711 |
{ |
3712 |
return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); |
3713 |
} |
3714 |
|
3715 |
static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
3716 |
{ |
3717 |
RwCo *rwco = opaque; |
3718 |
|
3719 |
rwco->ret = bdrv_co_flush(rwco->bs); |
3720 |
} |
3721 |
|
3722 |
int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
|
3723 |
{ |
3724 |
int ret;
|
3725 |
|
3726 |
if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
|
3727 |
return 0; |
3728 |
} |
3729 |
|
3730 |
/* Write back cached data to the OS even with cache=unsafe */
|
3731 |
if (bs->drv->bdrv_co_flush_to_os) {
|
3732 |
ret = bs->drv->bdrv_co_flush_to_os(bs); |
3733 |
if (ret < 0) { |
3734 |
return ret;
|
3735 |
} |
3736 |
} |
3737 |
|
3738 |
/* But don't actually force it to the disk with cache=unsafe */
|
3739 |
if (bs->open_flags & BDRV_O_NO_FLUSH) {
|
3740 |
goto flush_parent;
|
3741 |
} |
3742 |
|
3743 |
if (bs->drv->bdrv_co_flush_to_disk) {
|
3744 |
ret = bs->drv->bdrv_co_flush_to_disk(bs); |
3745 |
} else if (bs->drv->bdrv_aio_flush) { |
3746 |
BlockDriverAIOCB *acb; |
3747 |
CoroutineIOCompletion co = { |
3748 |
.coroutine = qemu_coroutine_self(), |
3749 |
}; |
3750 |
|
3751 |
acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); |
3752 |
if (acb == NULL) { |
3753 |
ret = -EIO; |
3754 |
} else {
|
3755 |
qemu_coroutine_yield(); |
3756 |
ret = co.ret; |
3757 |
} |
3758 |
} else {
|
3759 |
/*
|
3760 |
* Some block drivers always operate in either writethrough or unsafe
|
3761 |
* mode and don't support bdrv_flush therefore. Usually qemu doesn't
|
3762 |
* know how the server works (because the behaviour is hardcoded or
|
3763 |
* depends on server-side configuration), so we can't ensure that
|
3764 |
* everything is safe on disk. Returning an error doesn't work because
|
3765 |
* that would break guests even if the server operates in writethrough
|
3766 |
* mode.
|
3767 |
*
|
3768 |
* Let's hope the user knows what he's doing.
|
3769 |
*/
|
3770 |
ret = 0;
|
3771 |
} |
3772 |
if (ret < 0) { |
3773 |
return ret;
|
3774 |
} |
3775 |
|
3776 |
/* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
|
3777 |
* in the case of cache=unsafe, so there are no useless flushes.
|
3778 |
*/
|
3779 |
flush_parent:
|
3780 |
return bdrv_co_flush(bs->file);
|
3781 |
} |
3782 |
|
3783 |
void bdrv_invalidate_cache(BlockDriverState *bs)
|
3784 |
{ |
3785 |
if (bs->drv && bs->drv->bdrv_invalidate_cache) {
|
3786 |
bs->drv->bdrv_invalidate_cache(bs); |
3787 |
} |
3788 |
} |
3789 |
|
3790 |
void bdrv_invalidate_cache_all(void) |
3791 |
{ |
3792 |
BlockDriverState *bs; |
3793 |
|
3794 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
3795 |
bdrv_invalidate_cache(bs); |
3796 |
} |
3797 |
} |
3798 |
|
3799 |
void bdrv_clear_incoming_migration_all(void) |
3800 |
{ |
3801 |
BlockDriverState *bs; |
3802 |
|
3803 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
3804 |
bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING); |
3805 |
} |
3806 |
} |
3807 |
|
3808 |
int bdrv_flush(BlockDriverState *bs)
|
3809 |
{ |
3810 |
Coroutine *co; |
3811 |
RwCo rwco = { |
3812 |
.bs = bs, |
3813 |
.ret = NOT_DONE, |
3814 |
}; |
3815 |
|
3816 |
if (qemu_in_coroutine()) {
|
3817 |
/* Fast-path if already in coroutine context */
|
3818 |
bdrv_flush_co_entry(&rwco); |
3819 |
} else {
|
3820 |
co = qemu_coroutine_create(bdrv_flush_co_entry); |
3821 |
qemu_coroutine_enter(co, &rwco); |
3822 |
while (rwco.ret == NOT_DONE) {
|
3823 |
qemu_aio_wait(); |
3824 |
} |
3825 |
} |
3826 |
|
3827 |
return rwco.ret;
|
3828 |
} |
3829 |
|
3830 |
static void coroutine_fn bdrv_discard_co_entry(void *opaque) |
3831 |
{ |
3832 |
RwCo *rwco = opaque; |
3833 |
|
3834 |
rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); |
3835 |
} |
3836 |
|
3837 |
int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
|
3838 |
int nb_sectors)
|
3839 |
{ |
3840 |
if (!bs->drv) {
|
3841 |
return -ENOMEDIUM;
|
3842 |
} else if (bdrv_check_request(bs, sector_num, nb_sectors)) { |
3843 |
return -EIO;
|
3844 |
} else if (bs->read_only) { |
3845 |
return -EROFS;
|
3846 |
} else if (bs->drv->bdrv_co_discard) { |
3847 |
return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
|
3848 |
} else if (bs->drv->bdrv_aio_discard) { |
3849 |
BlockDriverAIOCB *acb; |
3850 |
CoroutineIOCompletion co = { |
3851 |
.coroutine = qemu_coroutine_self(), |
3852 |
}; |
3853 |
|
3854 |
acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, |
3855 |
bdrv_co_io_em_complete, &co); |
3856 |
if (acb == NULL) { |
3857 |
return -EIO;
|
3858 |
} else {
|
3859 |
qemu_coroutine_yield(); |
3860 |
return co.ret;
|
3861 |
} |
3862 |
} else {
|
3863 |
return 0; |
3864 |
} |
3865 |
} |
3866 |
|
3867 |
int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) |
3868 |
{ |
3869 |
Coroutine *co; |
3870 |
RwCo rwco = { |
3871 |
.bs = bs, |
3872 |
.sector_num = sector_num, |
3873 |
.nb_sectors = nb_sectors, |
3874 |
.ret = NOT_DONE, |
3875 |
}; |
3876 |
|
3877 |
if (qemu_in_coroutine()) {
|
3878 |
/* Fast-path if already in coroutine context */
|
3879 |
bdrv_discard_co_entry(&rwco); |
3880 |
} else {
|
3881 |
co = qemu_coroutine_create(bdrv_discard_co_entry); |
3882 |
qemu_coroutine_enter(co, &rwco); |
3883 |
while (rwco.ret == NOT_DONE) {
|
3884 |
qemu_aio_wait(); |
3885 |
} |
3886 |
} |
3887 |
|
3888 |
return rwco.ret;
|
3889 |
} |
3890 |
|
3891 |
/**************************************************************/
|
3892 |
/* removable device support */
|
3893 |
|
3894 |
/**
|
3895 |
* Return TRUE if the media is present
|
3896 |
*/
|
3897 |
int bdrv_is_inserted(BlockDriverState *bs)
|
3898 |
{ |
3899 |
BlockDriver *drv = bs->drv; |
3900 |
|
3901 |
if (!drv)
|
3902 |
return 0; |
3903 |
if (!drv->bdrv_is_inserted)
|
3904 |
return 1; |
3905 |
return drv->bdrv_is_inserted(bs);
|
3906 |
} |
3907 |
|
3908 |
/**
|
3909 |
* Return whether the media changed since the last call to this
|
3910 |
* function, or -ENOTSUP if we don't know. Most drivers don't know.
|
3911 |
*/
|
3912 |
int bdrv_media_changed(BlockDriverState *bs)
|
3913 |
{ |
3914 |
BlockDriver *drv = bs->drv; |
3915 |
|
3916 |
if (drv && drv->bdrv_media_changed) {
|
3917 |
return drv->bdrv_media_changed(bs);
|
3918 |
} |
3919 |
return -ENOTSUP;
|
3920 |
} |
3921 |
|
3922 |
/**
|
3923 |
* If eject_flag is TRUE, eject the media. Otherwise, close the tray
|
3924 |
*/
|
3925 |
void bdrv_eject(BlockDriverState *bs, bool eject_flag) |
3926 |
{ |
3927 |
BlockDriver *drv = bs->drv; |
3928 |
|
3929 |
if (drv && drv->bdrv_eject) {
|
3930 |
drv->bdrv_eject(bs, eject_flag); |
3931 |
} |
3932 |
|
3933 |
if (bs->device_name[0] != '\0') { |
3934 |
bdrv_emit_qmp_eject_event(bs, eject_flag); |
3935 |
} |
3936 |
} |
3937 |
|
3938 |
/**
|
3939 |
* Lock or unlock the media (if it is locked, the user won't be able
|
3940 |
* to eject it manually).
|
3941 |
*/
|
3942 |
void bdrv_lock_medium(BlockDriverState *bs, bool locked) |
3943 |
{ |
3944 |
BlockDriver *drv = bs->drv; |
3945 |
|
3946 |
trace_bdrv_lock_medium(bs, locked); |
3947 |
|
3948 |
if (drv && drv->bdrv_lock_medium) {
|
3949 |
drv->bdrv_lock_medium(bs, locked); |
3950 |
} |
3951 |
} |
3952 |
|
3953 |
/* needed for generic scsi interface */
|
3954 |
|
3955 |
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) |
3956 |
{ |
3957 |
BlockDriver *drv = bs->drv; |
3958 |
|
3959 |
if (drv && drv->bdrv_ioctl)
|
3960 |
return drv->bdrv_ioctl(bs, req, buf);
|
3961 |
return -ENOTSUP;
|
3962 |
} |
3963 |
|
3964 |
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, |
3965 |
unsigned long int req, void *buf, |
3966 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3967 |
{ |
3968 |
BlockDriver *drv = bs->drv; |
3969 |
|
3970 |
if (drv && drv->bdrv_aio_ioctl)
|
3971 |
return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
|
3972 |
return NULL; |
3973 |
} |
3974 |
|
3975 |
void bdrv_set_buffer_alignment(BlockDriverState *bs, int align) |
3976 |
{ |
3977 |
bs->buffer_alignment = align; |
3978 |
} |
3979 |
|
3980 |
void *qemu_blockalign(BlockDriverState *bs, size_t size)
|
3981 |
{ |
3982 |
return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size); |
3983 |
} |
3984 |
|
3985 |
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable) |
3986 |
{ |
3987 |
int64_t bitmap_size; |
3988 |
|
3989 |
bs->dirty_count = 0;
|
3990 |
if (enable) {
|
3991 |
if (!bs->dirty_bitmap) {
|
3992 |
bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) + |
3993 |
BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
|
3994 |
bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG; |
3995 |
|
3996 |
bs->dirty_bitmap = g_new0(unsigned long, bitmap_size); |
3997 |
} |
3998 |
} else {
|
3999 |
if (bs->dirty_bitmap) {
|
4000 |
g_free(bs->dirty_bitmap); |
4001 |
bs->dirty_bitmap = NULL;
|
4002 |
} |
4003 |
} |
4004 |
} |
4005 |
|
4006 |
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
|
4007 |
{ |
4008 |
int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; |
4009 |
|
4010 |
if (bs->dirty_bitmap &&
|
4011 |
(sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) { |
4012 |
return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] & |
4013 |
(1UL << (chunk % (sizeof(unsigned long) * 8)))); |
4014 |
} else {
|
4015 |
return 0; |
4016 |
} |
4017 |
} |
4018 |
|
4019 |
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
|
4020 |
int nr_sectors)
|
4021 |
{ |
4022 |
set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
|
4023 |
} |
4024 |
|
4025 |
int64_t bdrv_get_dirty_count(BlockDriverState *bs) |
4026 |
{ |
4027 |
return bs->dirty_count;
|
4028 |
} |
4029 |
|
4030 |
void bdrv_set_in_use(BlockDriverState *bs, int in_use) |
4031 |
{ |
4032 |
assert(bs->in_use != in_use); |
4033 |
bs->in_use = in_use; |
4034 |
} |
4035 |
|
4036 |
int bdrv_in_use(BlockDriverState *bs)
|
4037 |
{ |
4038 |
return bs->in_use;
|
4039 |
} |
4040 |
|
4041 |
void bdrv_iostatus_enable(BlockDriverState *bs)
|
4042 |
{ |
4043 |
bs->iostatus_enabled = true;
|
4044 |
bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; |
4045 |
} |
4046 |
|
4047 |
/* The I/O status is only enabled if the drive explicitly
|
4048 |
* enables it _and_ the VM is configured to stop on errors */
|
4049 |
bool bdrv_iostatus_is_enabled(const BlockDriverState *bs) |
4050 |
{ |
4051 |
return (bs->iostatus_enabled &&
|
4052 |
(bs->on_write_error == BLOCK_ERR_STOP_ENOSPC || |
4053 |
bs->on_write_error == BLOCK_ERR_STOP_ANY || |
4054 |
bs->on_read_error == BLOCK_ERR_STOP_ANY)); |
4055 |
} |
4056 |
|
4057 |
void bdrv_iostatus_disable(BlockDriverState *bs)
|
4058 |
{ |
4059 |
bs->iostatus_enabled = false;
|
4060 |
} |
4061 |
|
4062 |
void bdrv_iostatus_reset(BlockDriverState *bs)
|
4063 |
{ |
4064 |
if (bdrv_iostatus_is_enabled(bs)) {
|
4065 |
bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; |
4066 |
} |
4067 |
} |
4068 |
|
4069 |
/* XXX: Today this is set by device models because it makes the implementation
|
4070 |
quite simple. However, the block layer knows about the error, so it's
|
4071 |
possible to implement this without device models being involved */
|
4072 |
void bdrv_iostatus_set_err(BlockDriverState *bs, int error) |
4073 |
{ |
4074 |
if (bdrv_iostatus_is_enabled(bs) &&
|
4075 |
bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
4076 |
assert(error >= 0);
|
4077 |
bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : |
4078 |
BLOCK_DEVICE_IO_STATUS_FAILED; |
4079 |
} |
4080 |
} |
4081 |
|
4082 |
void
|
4083 |
bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes, |
4084 |
enum BlockAcctType type)
|
4085 |
{ |
4086 |
assert(type < BDRV_MAX_IOTYPE); |
4087 |
|
4088 |
cookie->bytes = bytes; |
4089 |
cookie->start_time_ns = get_clock(); |
4090 |
cookie->type = type; |
4091 |
} |
4092 |
|
4093 |
void
|
4094 |
bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie) |
4095 |
{ |
4096 |
assert(cookie->type < BDRV_MAX_IOTYPE); |
4097 |
|
4098 |
bs->nr_bytes[cookie->type] += cookie->bytes; |
4099 |
bs->nr_ops[cookie->type]++; |
4100 |
bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns; |
4101 |
} |
4102 |
|
4103 |
int bdrv_img_create(const char *filename, const char *fmt, |
4104 |
const char *base_filename, const char *base_fmt, |
4105 |
char *options, uint64_t img_size, int flags) |
4106 |
{ |
4107 |
QEMUOptionParameter *param = NULL, *create_options = NULL; |
4108 |
QEMUOptionParameter *backing_fmt, *backing_file, *size; |
4109 |
BlockDriverState *bs = NULL;
|
4110 |
BlockDriver *drv, *proto_drv; |
4111 |
BlockDriver *backing_drv = NULL;
|
4112 |
int ret = 0; |
4113 |
|
4114 |
/* Find driver and parse its options */
|
4115 |
drv = bdrv_find_format(fmt); |
4116 |
if (!drv) {
|
4117 |
error_report("Unknown file format '%s'", fmt);
|
4118 |
ret = -EINVAL; |
4119 |
goto out;
|
4120 |
} |
4121 |
|
4122 |
proto_drv = bdrv_find_protocol(filename); |
4123 |
if (!proto_drv) {
|
4124 |
error_report("Unknown protocol '%s'", filename);
|
4125 |
ret = -EINVAL; |
4126 |
goto out;
|
4127 |
} |
4128 |
|
4129 |
create_options = append_option_parameters(create_options, |
4130 |
drv->create_options); |
4131 |
create_options = append_option_parameters(create_options, |
4132 |
proto_drv->create_options); |
4133 |
|
4134 |
/* Create parameter list with default values */
|
4135 |
param = parse_option_parameters("", create_options, param);
|
4136 |
|
4137 |
set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size); |
4138 |
|
4139 |
/* Parse -o options */
|
4140 |
if (options) {
|
4141 |
param = parse_option_parameters(options, create_options, param); |
4142 |
if (param == NULL) { |
4143 |
error_report("Invalid options for file format '%s'.", fmt);
|
4144 |
ret = -EINVAL; |
4145 |
goto out;
|
4146 |
} |
4147 |
} |
4148 |
|
4149 |
if (base_filename) {
|
4150 |
if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
|
4151 |
base_filename)) { |
4152 |
error_report("Backing file not supported for file format '%s'",
|
4153 |
fmt); |
4154 |
ret = -EINVAL; |
4155 |
goto out;
|
4156 |
} |
4157 |
} |
4158 |
|
4159 |
if (base_fmt) {
|
4160 |
if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
|
4161 |
error_report("Backing file format not supported for file "
|
4162 |
"format '%s'", fmt);
|
4163 |
ret = -EINVAL; |
4164 |
goto out;
|
4165 |
} |
4166 |
} |
4167 |
|
4168 |
backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE); |
4169 |
if (backing_file && backing_file->value.s) {
|
4170 |
if (!strcmp(filename, backing_file->value.s)) {
|
4171 |
error_report("Error: Trying to create an image with the "
|
4172 |
"same filename as the backing file");
|
4173 |
ret = -EINVAL; |
4174 |
goto out;
|
4175 |
} |
4176 |
} |
4177 |
|
4178 |
backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT); |
4179 |
if (backing_fmt && backing_fmt->value.s) {
|
4180 |
backing_drv = bdrv_find_format(backing_fmt->value.s); |
4181 |
if (!backing_drv) {
|
4182 |
error_report("Unknown backing file format '%s'",
|
4183 |
backing_fmt->value.s); |
4184 |
ret = -EINVAL; |
4185 |
goto out;
|
4186 |
} |
4187 |
} |
4188 |
|
4189 |
// The size for the image must always be specified, with one exception:
|
4190 |
// If we are using a backing file, we can obtain the size from there
|
4191 |
size = get_option_parameter(param, BLOCK_OPT_SIZE); |
4192 |
if (size && size->value.n == -1) { |
4193 |
if (backing_file && backing_file->value.s) {
|
4194 |
uint64_t size; |
4195 |
char buf[32]; |
4196 |
int back_flags;
|
4197 |
|
4198 |
/* backing files always opened read-only */
|
4199 |
back_flags = |
4200 |
flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
4201 |
|
4202 |
bs = bdrv_new("");
|
4203 |
|
4204 |
ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv); |
4205 |
if (ret < 0) { |
4206 |
error_report("Could not open '%s'", backing_file->value.s);
|
4207 |
goto out;
|
4208 |
} |
4209 |
bdrv_get_geometry(bs, &size); |
4210 |
size *= 512;
|
4211 |
|
4212 |
snprintf(buf, sizeof(buf), "%" PRId64, size); |
4213 |
set_option_parameter(param, BLOCK_OPT_SIZE, buf); |
4214 |
} else {
|
4215 |
error_report("Image creation needs a size parameter");
|
4216 |
ret = -EINVAL; |
4217 |
goto out;
|
4218 |
} |
4219 |
} |
4220 |
|
4221 |
printf("Formatting '%s', fmt=%s ", filename, fmt);
|
4222 |
print_option_parameters(param); |
4223 |
puts("");
|
4224 |
|
4225 |
ret = bdrv_create(drv, filename, param); |
4226 |
|
4227 |
if (ret < 0) { |
4228 |
if (ret == -ENOTSUP) {
|
4229 |
error_report("Formatting or formatting option not supported for "
|
4230 |
"file format '%s'", fmt);
|
4231 |
} else if (ret == -EFBIG) { |
4232 |
error_report("The image size is too large for file format '%s'",
|
4233 |
fmt); |
4234 |
} else {
|
4235 |
error_report("%s: error while creating %s: %s", filename, fmt,
|
4236 |
strerror(-ret)); |
4237 |
} |
4238 |
} |
4239 |
|
4240 |
out:
|
4241 |
free_option_parameters(create_options); |
4242 |
free_option_parameters(param); |
4243 |
|
4244 |
if (bs) {
|
4245 |
bdrv_delete(bs); |
4246 |
} |
4247 |
|
4248 |
return ret;
|
4249 |
} |
4250 |
|
4251 |
void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs, |
4252 |
int64_t speed, BlockDriverCompletionFunc *cb, |
4253 |
void *opaque, Error **errp)
|
4254 |
{ |
4255 |
BlockJob *job; |
4256 |
|
4257 |
if (bs->job || bdrv_in_use(bs)) {
|
4258 |
error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs)); |
4259 |
return NULL; |
4260 |
} |
4261 |
bdrv_set_in_use(bs, 1);
|
4262 |
|
4263 |
job = g_malloc0(job_type->instance_size); |
4264 |
job->job_type = job_type; |
4265 |
job->bs = bs; |
4266 |
job->cb = cb; |
4267 |
job->opaque = opaque; |
4268 |
job->busy = true;
|
4269 |
bs->job = job; |
4270 |
|
4271 |
/* Only set speed when necessary to avoid NotSupported error */
|
4272 |
if (speed != 0) { |
4273 |
Error *local_err = NULL;
|
4274 |
|
4275 |
block_job_set_speed(job, speed, &local_err); |
4276 |
if (error_is_set(&local_err)) {
|
4277 |
bs->job = NULL;
|
4278 |
g_free(job); |
4279 |
bdrv_set_in_use(bs, 0);
|
4280 |
error_propagate(errp, local_err); |
4281 |
return NULL; |
4282 |
} |
4283 |
} |
4284 |
return job;
|
4285 |
} |
4286 |
|
4287 |
void block_job_complete(BlockJob *job, int ret) |
4288 |
{ |
4289 |
BlockDriverState *bs = job->bs; |
4290 |
|
4291 |
assert(bs->job == job); |
4292 |
job->cb(job->opaque, ret); |
4293 |
bs->job = NULL;
|
4294 |
g_free(job); |
4295 |
bdrv_set_in_use(bs, 0);
|
4296 |
} |
4297 |
|
4298 |
void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
|
4299 |
{ |
4300 |
Error *local_err = NULL;
|
4301 |
|
4302 |
if (!job->job_type->set_speed) {
|
4303 |
error_set(errp, QERR_NOT_SUPPORTED); |
4304 |
return;
|
4305 |
} |
4306 |
job->job_type->set_speed(job, speed, &local_err); |
4307 |
if (error_is_set(&local_err)) {
|
4308 |
error_propagate(errp, local_err); |
4309 |
return;
|
4310 |
} |
4311 |
|
4312 |
job->speed = speed; |
4313 |
} |
4314 |
|
4315 |
void block_job_cancel(BlockJob *job)
|
4316 |
{ |
4317 |
job->cancelled = true;
|
4318 |
if (job->co && !job->busy) {
|
4319 |
qemu_coroutine_enter(job->co, NULL);
|
4320 |
} |
4321 |
} |
4322 |
|
4323 |
bool block_job_is_cancelled(BlockJob *job)
|
4324 |
{ |
4325 |
return job->cancelled;
|
4326 |
} |
4327 |
|
4328 |
struct BlockCancelData {
|
4329 |
BlockJob *job; |
4330 |
BlockDriverCompletionFunc *cb; |
4331 |
void *opaque;
|
4332 |
bool cancelled;
|
4333 |
int ret;
|
4334 |
}; |
4335 |
|
4336 |
static void block_job_cancel_cb(void *opaque, int ret) |
4337 |
{ |
4338 |
struct BlockCancelData *data = opaque;
|
4339 |
|
4340 |
data->cancelled = block_job_is_cancelled(data->job); |
4341 |
data->ret = ret; |
4342 |
data->cb(data->opaque, ret); |
4343 |
} |
4344 |
|
4345 |
int block_job_cancel_sync(BlockJob *job)
|
4346 |
{ |
4347 |
struct BlockCancelData data;
|
4348 |
BlockDriverState *bs = job->bs; |
4349 |
|
4350 |
assert(bs->job == job); |
4351 |
|
4352 |
/* Set up our own callback to store the result and chain to
|
4353 |
* the original callback.
|
4354 |
*/
|
4355 |
data.job = job; |
4356 |
data.cb = job->cb; |
4357 |
data.opaque = job->opaque; |
4358 |
data.ret = -EINPROGRESS; |
4359 |
job->cb = block_job_cancel_cb; |
4360 |
job->opaque = &data; |
4361 |
block_job_cancel(job); |
4362 |
while (data.ret == -EINPROGRESS) {
|
4363 |
qemu_aio_wait(); |
4364 |
} |
4365 |
return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret; |
4366 |
} |
4367 |
|
4368 |
void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
|
4369 |
{ |
4370 |
/* Check cancellation *before* setting busy = false, too! */
|
4371 |
if (!block_job_is_cancelled(job)) {
|
4372 |
job->busy = false;
|
4373 |
co_sleep_ns(clock, ns); |
4374 |
job->busy = true;
|
4375 |
} |
4376 |
} |