root / block.c @ e1e9b0ac
History | View | Annotate | Download (120.9 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 || close(fd)) { |
437 |
return -errno;
|
438 |
} |
439 |
return 0; |
440 |
#endif
|
441 |
} |
442 |
|
443 |
/*
|
444 |
* Detect host devices. By convention, /dev/cdrom[N] is always
|
445 |
* recognized as a host CDROM.
|
446 |
*/
|
447 |
static BlockDriver *find_hdev_driver(const char *filename) |
448 |
{ |
449 |
int score_max = 0, score; |
450 |
BlockDriver *drv = NULL, *d;
|
451 |
|
452 |
QLIST_FOREACH(d, &bdrv_drivers, list) { |
453 |
if (d->bdrv_probe_device) {
|
454 |
score = d->bdrv_probe_device(filename); |
455 |
if (score > score_max) {
|
456 |
score_max = score; |
457 |
drv = d; |
458 |
} |
459 |
} |
460 |
} |
461 |
|
462 |
return drv;
|
463 |
} |
464 |
|
465 |
BlockDriver *bdrv_find_protocol(const char *filename) |
466 |
{ |
467 |
BlockDriver *drv1; |
468 |
char protocol[128]; |
469 |
int len;
|
470 |
const char *p; |
471 |
|
472 |
/* TODO Drivers without bdrv_file_open must be specified explicitly */
|
473 |
|
474 |
/*
|
475 |
* XXX(hch): we really should not let host device detection
|
476 |
* override an explicit protocol specification, but moving this
|
477 |
* later breaks access to device names with colons in them.
|
478 |
* Thanks to the brain-dead persistent naming schemes on udev-
|
479 |
* based Linux systems those actually are quite common.
|
480 |
*/
|
481 |
drv1 = find_hdev_driver(filename); |
482 |
if (drv1) {
|
483 |
return drv1;
|
484 |
} |
485 |
|
486 |
if (!path_has_protocol(filename)) {
|
487 |
return bdrv_find_format("file"); |
488 |
} |
489 |
p = strchr(filename, ':');
|
490 |
assert(p != NULL);
|
491 |
len = p - filename; |
492 |
if (len > sizeof(protocol) - 1) |
493 |
len = sizeof(protocol) - 1; |
494 |
memcpy(protocol, filename, len); |
495 |
protocol[len] = '\0';
|
496 |
QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
497 |
if (drv1->protocol_name &&
|
498 |
!strcmp(drv1->protocol_name, protocol)) { |
499 |
return drv1;
|
500 |
} |
501 |
} |
502 |
return NULL; |
503 |
} |
504 |
|
505 |
static int find_image_format(const char *filename, BlockDriver **pdrv) |
506 |
{ |
507 |
int ret, score, score_max;
|
508 |
BlockDriver *drv1, *drv; |
509 |
uint8_t buf[2048];
|
510 |
BlockDriverState *bs; |
511 |
|
512 |
ret = bdrv_file_open(&bs, filename, 0);
|
513 |
if (ret < 0) { |
514 |
*pdrv = NULL;
|
515 |
return ret;
|
516 |
} |
517 |
|
518 |
/* Return the raw BlockDriver * to scsi-generic devices or empty drives */
|
519 |
if (bs->sg || !bdrv_is_inserted(bs)) {
|
520 |
bdrv_delete(bs); |
521 |
drv = bdrv_find_format("raw");
|
522 |
if (!drv) {
|
523 |
ret = -ENOENT; |
524 |
} |
525 |
*pdrv = drv; |
526 |
return ret;
|
527 |
} |
528 |
|
529 |
ret = bdrv_pread(bs, 0, buf, sizeof(buf)); |
530 |
bdrv_delete(bs); |
531 |
if (ret < 0) { |
532 |
*pdrv = NULL;
|
533 |
return ret;
|
534 |
} |
535 |
|
536 |
score_max = 0;
|
537 |
drv = NULL;
|
538 |
QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
539 |
if (drv1->bdrv_probe) {
|
540 |
score = drv1->bdrv_probe(buf, ret, filename); |
541 |
if (score > score_max) {
|
542 |
score_max = score; |
543 |
drv = drv1; |
544 |
} |
545 |
} |
546 |
} |
547 |
if (!drv) {
|
548 |
ret = -ENOENT; |
549 |
} |
550 |
*pdrv = drv; |
551 |
return ret;
|
552 |
} |
553 |
|
554 |
/**
|
555 |
* Set the current 'total_sectors' value
|
556 |
*/
|
557 |
static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) |
558 |
{ |
559 |
BlockDriver *drv = bs->drv; |
560 |
|
561 |
/* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
|
562 |
if (bs->sg)
|
563 |
return 0; |
564 |
|
565 |
/* query actual device if possible, otherwise just trust the hint */
|
566 |
if (drv->bdrv_getlength) {
|
567 |
int64_t length = drv->bdrv_getlength(bs); |
568 |
if (length < 0) { |
569 |
return length;
|
570 |
} |
571 |
hint = length >> BDRV_SECTOR_BITS; |
572 |
} |
573 |
|
574 |
bs->total_sectors = hint; |
575 |
return 0; |
576 |
} |
577 |
|
578 |
/**
|
579 |
* Set open flags for a given cache mode
|
580 |
*
|
581 |
* Return 0 on success, -1 if the cache mode was invalid.
|
582 |
*/
|
583 |
int bdrv_parse_cache_flags(const char *mode, int *flags) |
584 |
{ |
585 |
*flags &= ~BDRV_O_CACHE_MASK; |
586 |
|
587 |
if (!strcmp(mode, "off") || !strcmp(mode, "none")) { |
588 |
*flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB; |
589 |
} else if (!strcmp(mode, "directsync")) { |
590 |
*flags |= BDRV_O_NOCACHE; |
591 |
} else if (!strcmp(mode, "writeback")) { |
592 |
*flags |= BDRV_O_CACHE_WB; |
593 |
} else if (!strcmp(mode, "unsafe")) { |
594 |
*flags |= BDRV_O_CACHE_WB; |
595 |
*flags |= BDRV_O_NO_FLUSH; |
596 |
} else if (!strcmp(mode, "writethrough")) { |
597 |
/* this is the default */
|
598 |
} else {
|
599 |
return -1; |
600 |
} |
601 |
|
602 |
return 0; |
603 |
} |
604 |
|
605 |
/**
|
606 |
* The copy-on-read flag is actually a reference count so multiple users may
|
607 |
* use the feature without worrying about clobbering its previous state.
|
608 |
* Copy-on-read stays enabled until all users have called to disable it.
|
609 |
*/
|
610 |
void bdrv_enable_copy_on_read(BlockDriverState *bs)
|
611 |
{ |
612 |
bs->copy_on_read++; |
613 |
} |
614 |
|
615 |
void bdrv_disable_copy_on_read(BlockDriverState *bs)
|
616 |
{ |
617 |
assert(bs->copy_on_read > 0);
|
618 |
bs->copy_on_read--; |
619 |
} |
620 |
|
621 |
/*
|
622 |
* Common part for opening disk images and files
|
623 |
*/
|
624 |
static int bdrv_open_common(BlockDriverState *bs, const char *filename, |
625 |
int flags, BlockDriver *drv)
|
626 |
{ |
627 |
int ret, open_flags;
|
628 |
|
629 |
assert(drv != NULL);
|
630 |
assert(bs->file == NULL);
|
631 |
|
632 |
trace_bdrv_open_common(bs, filename, flags, drv->format_name); |
633 |
|
634 |
bs->open_flags = flags; |
635 |
bs->buffer_alignment = 512;
|
636 |
|
637 |
assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ |
638 |
if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
|
639 |
bdrv_enable_copy_on_read(bs); |
640 |
} |
641 |
|
642 |
pstrcpy(bs->filename, sizeof(bs->filename), filename);
|
643 |
|
644 |
if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
|
645 |
return -ENOTSUP;
|
646 |
} |
647 |
|
648 |
bs->drv = drv; |
649 |
bs->opaque = g_malloc0(drv->instance_size); |
650 |
|
651 |
bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB); |
652 |
open_flags = flags | BDRV_O_CACHE_WB; |
653 |
|
654 |
/*
|
655 |
* Clear flags that are internal to the block layer before opening the
|
656 |
* image.
|
657 |
*/
|
658 |
open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
659 |
|
660 |
/*
|
661 |
* Snapshots should be writable.
|
662 |
*/
|
663 |
if (bs->is_temporary) {
|
664 |
open_flags |= BDRV_O_RDWR; |
665 |
} |
666 |
|
667 |
bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR); |
668 |
|
669 |
/* Open the image, either directly or using a protocol */
|
670 |
if (drv->bdrv_file_open) {
|
671 |
ret = drv->bdrv_file_open(bs, filename, open_flags); |
672 |
} else {
|
673 |
ret = bdrv_file_open(&bs->file, filename, open_flags); |
674 |
if (ret >= 0) { |
675 |
ret = drv->bdrv_open(bs, open_flags); |
676 |
} |
677 |
} |
678 |
|
679 |
if (ret < 0) { |
680 |
goto free_and_fail;
|
681 |
} |
682 |
|
683 |
ret = refresh_total_sectors(bs, bs->total_sectors); |
684 |
if (ret < 0) { |
685 |
goto free_and_fail;
|
686 |
} |
687 |
|
688 |
#ifndef _WIN32
|
689 |
if (bs->is_temporary) {
|
690 |
unlink(filename); |
691 |
} |
692 |
#endif
|
693 |
return 0; |
694 |
|
695 |
free_and_fail:
|
696 |
if (bs->file) {
|
697 |
bdrv_delete(bs->file); |
698 |
bs->file = NULL;
|
699 |
} |
700 |
g_free(bs->opaque); |
701 |
bs->opaque = NULL;
|
702 |
bs->drv = NULL;
|
703 |
return ret;
|
704 |
} |
705 |
|
706 |
/*
|
707 |
* Opens a file using a protocol (file, host_device, nbd, ...)
|
708 |
*/
|
709 |
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags) |
710 |
{ |
711 |
BlockDriverState *bs; |
712 |
BlockDriver *drv; |
713 |
int ret;
|
714 |
|
715 |
drv = bdrv_find_protocol(filename); |
716 |
if (!drv) {
|
717 |
return -ENOENT;
|
718 |
} |
719 |
|
720 |
bs = bdrv_new("");
|
721 |
ret = bdrv_open_common(bs, filename, flags, drv); |
722 |
if (ret < 0) { |
723 |
bdrv_delete(bs); |
724 |
return ret;
|
725 |
} |
726 |
bs->growable = 1;
|
727 |
*pbs = bs; |
728 |
return 0; |
729 |
} |
730 |
|
731 |
/*
|
732 |
* Opens a disk image (raw, qcow2, vmdk, ...)
|
733 |
*/
|
734 |
int bdrv_open(BlockDriverState *bs, const char *filename, int flags, |
735 |
BlockDriver *drv) |
736 |
{ |
737 |
int ret;
|
738 |
char tmp_filename[PATH_MAX];
|
739 |
|
740 |
if (flags & BDRV_O_SNAPSHOT) {
|
741 |
BlockDriverState *bs1; |
742 |
int64_t total_size; |
743 |
int is_protocol = 0; |
744 |
BlockDriver *bdrv_qcow2; |
745 |
QEMUOptionParameter *options; |
746 |
char backing_filename[PATH_MAX];
|
747 |
|
748 |
/* if snapshot, we create a temporary backing file and open it
|
749 |
instead of opening 'filename' directly */
|
750 |
|
751 |
/* if there is a backing file, use it */
|
752 |
bs1 = bdrv_new("");
|
753 |
ret = bdrv_open(bs1, filename, 0, drv);
|
754 |
if (ret < 0) { |
755 |
bdrv_delete(bs1); |
756 |
return ret;
|
757 |
} |
758 |
total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK; |
759 |
|
760 |
if (bs1->drv && bs1->drv->protocol_name)
|
761 |
is_protocol = 1;
|
762 |
|
763 |
bdrv_delete(bs1); |
764 |
|
765 |
ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
|
766 |
if (ret < 0) { |
767 |
return ret;
|
768 |
} |
769 |
|
770 |
/* Real path is meaningless for protocols */
|
771 |
if (is_protocol)
|
772 |
snprintf(backing_filename, sizeof(backing_filename),
|
773 |
"%s", filename);
|
774 |
else if (!realpath(filename, backing_filename)) |
775 |
return -errno;
|
776 |
|
777 |
bdrv_qcow2 = bdrv_find_format("qcow2");
|
778 |
options = parse_option_parameters("", bdrv_qcow2->create_options, NULL); |
779 |
|
780 |
set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size); |
781 |
set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename); |
782 |
if (drv) {
|
783 |
set_option_parameter(options, BLOCK_OPT_BACKING_FMT, |
784 |
drv->format_name); |
785 |
} |
786 |
|
787 |
ret = bdrv_create(bdrv_qcow2, tmp_filename, options); |
788 |
free_option_parameters(options); |
789 |
if (ret < 0) { |
790 |
return ret;
|
791 |
} |
792 |
|
793 |
filename = tmp_filename; |
794 |
drv = bdrv_qcow2; |
795 |
bs->is_temporary = 1;
|
796 |
} |
797 |
|
798 |
/* Find the right image format driver */
|
799 |
if (!drv) {
|
800 |
ret = find_image_format(filename, &drv); |
801 |
} |
802 |
|
803 |
if (!drv) {
|
804 |
goto unlink_and_fail;
|
805 |
} |
806 |
|
807 |
/* Open the image */
|
808 |
ret = bdrv_open_common(bs, filename, flags, drv); |
809 |
if (ret < 0) { |
810 |
goto unlink_and_fail;
|
811 |
} |
812 |
|
813 |
/* If there is a backing file, use it */
|
814 |
if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') { |
815 |
char backing_filename[PATH_MAX];
|
816 |
int back_flags;
|
817 |
BlockDriver *back_drv = NULL;
|
818 |
|
819 |
bs->backing_hd = bdrv_new("");
|
820 |
bdrv_get_full_backing_filename(bs, backing_filename, |
821 |
sizeof(backing_filename));
|
822 |
|
823 |
if (bs->backing_format[0] != '\0') { |
824 |
back_drv = bdrv_find_format(bs->backing_format); |
825 |
} |
826 |
|
827 |
/* backing files always opened read-only */
|
828 |
back_flags = |
829 |
flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
830 |
|
831 |
ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv); |
832 |
if (ret < 0) { |
833 |
bdrv_close(bs); |
834 |
return ret;
|
835 |
} |
836 |
if (bs->is_temporary) {
|
837 |
bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR); |
838 |
} else {
|
839 |
/* base image inherits from "parent" */
|
840 |
bs->backing_hd->keep_read_only = bs->keep_read_only; |
841 |
} |
842 |
} |
843 |
|
844 |
if (!bdrv_key_required(bs)) {
|
845 |
bdrv_dev_change_media_cb(bs, true);
|
846 |
} |
847 |
|
848 |
/* throttling disk I/O limits */
|
849 |
if (bs->io_limits_enabled) {
|
850 |
bdrv_io_limits_enable(bs); |
851 |
} |
852 |
|
853 |
return 0; |
854 |
|
855 |
unlink_and_fail:
|
856 |
if (bs->is_temporary) {
|
857 |
unlink(filename); |
858 |
} |
859 |
return ret;
|
860 |
} |
861 |
|
862 |
void bdrv_close(BlockDriverState *bs)
|
863 |
{ |
864 |
bdrv_flush(bs); |
865 |
if (bs->drv) {
|
866 |
if (bs->job) {
|
867 |
block_job_cancel_sync(bs->job); |
868 |
} |
869 |
bdrv_drain_all(); |
870 |
|
871 |
if (bs == bs_snapshots) {
|
872 |
bs_snapshots = NULL;
|
873 |
} |
874 |
if (bs->backing_hd) {
|
875 |
bdrv_delete(bs->backing_hd); |
876 |
bs->backing_hd = NULL;
|
877 |
} |
878 |
bs->drv->bdrv_close(bs); |
879 |
g_free(bs->opaque); |
880 |
#ifdef _WIN32
|
881 |
if (bs->is_temporary) {
|
882 |
unlink(bs->filename); |
883 |
} |
884 |
#endif
|
885 |
bs->opaque = NULL;
|
886 |
bs->drv = NULL;
|
887 |
bs->copy_on_read = 0;
|
888 |
bs->backing_file[0] = '\0'; |
889 |
bs->backing_format[0] = '\0'; |
890 |
bs->total_sectors = 0;
|
891 |
bs->encrypted = 0;
|
892 |
bs->valid_key = 0;
|
893 |
bs->sg = 0;
|
894 |
bs->growable = 0;
|
895 |
|
896 |
if (bs->file != NULL) { |
897 |
bdrv_delete(bs->file); |
898 |
bs->file = NULL;
|
899 |
} |
900 |
|
901 |
bdrv_dev_change_media_cb(bs, false);
|
902 |
} |
903 |
|
904 |
/*throttling disk I/O limits*/
|
905 |
if (bs->io_limits_enabled) {
|
906 |
bdrv_io_limits_disable(bs); |
907 |
} |
908 |
} |
909 |
|
910 |
void bdrv_close_all(void) |
911 |
{ |
912 |
BlockDriverState *bs; |
913 |
|
914 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
915 |
bdrv_close(bs); |
916 |
} |
917 |
} |
918 |
|
919 |
/*
|
920 |
* Wait for pending requests to complete across all BlockDriverStates
|
921 |
*
|
922 |
* This function does not flush data to disk, use bdrv_flush_all() for that
|
923 |
* after calling this function.
|
924 |
*
|
925 |
* Note that completion of an asynchronous I/O operation can trigger any
|
926 |
* number of other I/O operations on other devices---for example a coroutine
|
927 |
* can be arbitrarily complex and a constant flow of I/O can come until the
|
928 |
* coroutine is complete. Because of this, it is not possible to have a
|
929 |
* function to drain a single device's I/O queue.
|
930 |
*/
|
931 |
void bdrv_drain_all(void) |
932 |
{ |
933 |
BlockDriverState *bs; |
934 |
bool busy;
|
935 |
|
936 |
do {
|
937 |
busy = qemu_aio_wait(); |
938 |
|
939 |
/* FIXME: We do not have timer support here, so this is effectively
|
940 |
* a busy wait.
|
941 |
*/
|
942 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
943 |
if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
|
944 |
qemu_co_queue_restart_all(&bs->throttled_reqs); |
945 |
busy = true;
|
946 |
} |
947 |
} |
948 |
} while (busy);
|
949 |
|
950 |
/* If requests are still pending there is a bug somewhere */
|
951 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
952 |
assert(QLIST_EMPTY(&bs->tracked_requests)); |
953 |
assert(qemu_co_queue_empty(&bs->throttled_reqs)); |
954 |
} |
955 |
} |
956 |
|
957 |
/* make a BlockDriverState anonymous by removing from bdrv_state list.
|
958 |
Also, NULL terminate the device_name to prevent double remove */
|
959 |
void bdrv_make_anon(BlockDriverState *bs)
|
960 |
{ |
961 |
if (bs->device_name[0] != '\0') { |
962 |
QTAILQ_REMOVE(&bdrv_states, bs, list); |
963 |
} |
964 |
bs->device_name[0] = '\0'; |
965 |
} |
966 |
|
967 |
static void bdrv_rebind(BlockDriverState *bs) |
968 |
{ |
969 |
if (bs->drv && bs->drv->bdrv_rebind) {
|
970 |
bs->drv->bdrv_rebind(bs); |
971 |
} |
972 |
} |
973 |
|
974 |
/*
|
975 |
* Add new bs contents at the top of an image chain while the chain is
|
976 |
* live, while keeping required fields on the top layer.
|
977 |
*
|
978 |
* This will modify the BlockDriverState fields, and swap contents
|
979 |
* between bs_new and bs_top. Both bs_new and bs_top are modified.
|
980 |
*
|
981 |
* bs_new is required to be anonymous.
|
982 |
*
|
983 |
* This function does not create any image files.
|
984 |
*/
|
985 |
void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
|
986 |
{ |
987 |
BlockDriverState tmp; |
988 |
|
989 |
/* bs_new must be anonymous */
|
990 |
assert(bs_new->device_name[0] == '\0'); |
991 |
|
992 |
tmp = *bs_new; |
993 |
|
994 |
/* there are some fields that need to stay on the top layer: */
|
995 |
tmp.open_flags = bs_top->open_flags; |
996 |
|
997 |
/* dev info */
|
998 |
tmp.dev_ops = bs_top->dev_ops; |
999 |
tmp.dev_opaque = bs_top->dev_opaque; |
1000 |
tmp.dev = bs_top->dev; |
1001 |
tmp.buffer_alignment = bs_top->buffer_alignment; |
1002 |
tmp.copy_on_read = bs_top->copy_on_read; |
1003 |
|
1004 |
tmp.enable_write_cache = bs_top->enable_write_cache; |
1005 |
|
1006 |
/* i/o timing parameters */
|
1007 |
tmp.slice_time = bs_top->slice_time; |
1008 |
tmp.slice_start = bs_top->slice_start; |
1009 |
tmp.slice_end = bs_top->slice_end; |
1010 |
tmp.io_limits = bs_top->io_limits; |
1011 |
tmp.io_base = bs_top->io_base; |
1012 |
tmp.throttled_reqs = bs_top->throttled_reqs; |
1013 |
tmp.block_timer = bs_top->block_timer; |
1014 |
tmp.io_limits_enabled = bs_top->io_limits_enabled; |
1015 |
|
1016 |
/* geometry */
|
1017 |
tmp.cyls = bs_top->cyls; |
1018 |
tmp.heads = bs_top->heads; |
1019 |
tmp.secs = bs_top->secs; |
1020 |
tmp.translation = bs_top->translation; |
1021 |
|
1022 |
/* r/w error */
|
1023 |
tmp.on_read_error = bs_top->on_read_error; |
1024 |
tmp.on_write_error = bs_top->on_write_error; |
1025 |
|
1026 |
/* i/o status */
|
1027 |
tmp.iostatus_enabled = bs_top->iostatus_enabled; |
1028 |
tmp.iostatus = bs_top->iostatus; |
1029 |
|
1030 |
/* keep the same entry in bdrv_states */
|
1031 |
pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
|
1032 |
tmp.list = bs_top->list; |
1033 |
|
1034 |
/* The contents of 'tmp' will become bs_top, as we are
|
1035 |
* swapping bs_new and bs_top contents. */
|
1036 |
tmp.backing_hd = bs_new; |
1037 |
pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
|
1038 |
bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
|
1039 |
|
1040 |
/* swap contents of the fixed new bs and the current top */
|
1041 |
*bs_new = *bs_top; |
1042 |
*bs_top = tmp; |
1043 |
|
1044 |
/* device_name[] was carried over from the old bs_top. bs_new
|
1045 |
* shouldn't be in bdrv_states, so we need to make device_name[]
|
1046 |
* reflect the anonymity of bs_new
|
1047 |
*/
|
1048 |
bs_new->device_name[0] = '\0'; |
1049 |
|
1050 |
/* clear the copied fields in the new backing file */
|
1051 |
bdrv_detach_dev(bs_new, bs_new->dev); |
1052 |
|
1053 |
qemu_co_queue_init(&bs_new->throttled_reqs); |
1054 |
memset(&bs_new->io_base, 0, sizeof(bs_new->io_base)); |
1055 |
memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits)); |
1056 |
bdrv_iostatus_disable(bs_new); |
1057 |
|
1058 |
/* we don't use bdrv_io_limits_disable() for this, because we don't want
|
1059 |
* to affect or delete the block_timer, as it has been moved to bs_top */
|
1060 |
bs_new->io_limits_enabled = false;
|
1061 |
bs_new->block_timer = NULL;
|
1062 |
bs_new->slice_time = 0;
|
1063 |
bs_new->slice_start = 0;
|
1064 |
bs_new->slice_end = 0;
|
1065 |
|
1066 |
bdrv_rebind(bs_new); |
1067 |
bdrv_rebind(bs_top); |
1068 |
} |
1069 |
|
1070 |
void bdrv_delete(BlockDriverState *bs)
|
1071 |
{ |
1072 |
assert(!bs->dev); |
1073 |
assert(!bs->job); |
1074 |
assert(!bs->in_use); |
1075 |
|
1076 |
/* remove from list, if necessary */
|
1077 |
bdrv_make_anon(bs); |
1078 |
|
1079 |
bdrv_close(bs); |
1080 |
|
1081 |
assert(bs != bs_snapshots); |
1082 |
g_free(bs); |
1083 |
} |
1084 |
|
1085 |
int bdrv_attach_dev(BlockDriverState *bs, void *dev) |
1086 |
/* TODO change to DeviceState *dev when all users are qdevified */
|
1087 |
{ |
1088 |
if (bs->dev) {
|
1089 |
return -EBUSY;
|
1090 |
} |
1091 |
bs->dev = dev; |
1092 |
bdrv_iostatus_reset(bs); |
1093 |
return 0; |
1094 |
} |
1095 |
|
1096 |
/* TODO qdevified devices don't use this, remove when devices are qdevified */
|
1097 |
void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev) |
1098 |
{ |
1099 |
if (bdrv_attach_dev(bs, dev) < 0) { |
1100 |
abort(); |
1101 |
} |
1102 |
} |
1103 |
|
1104 |
void bdrv_detach_dev(BlockDriverState *bs, void *dev) |
1105 |
/* TODO change to DeviceState *dev when all users are qdevified */
|
1106 |
{ |
1107 |
assert(bs->dev == dev); |
1108 |
bs->dev = NULL;
|
1109 |
bs->dev_ops = NULL;
|
1110 |
bs->dev_opaque = NULL;
|
1111 |
bs->buffer_alignment = 512;
|
1112 |
} |
1113 |
|
1114 |
/* TODO change to return DeviceState * when all users are qdevified */
|
1115 |
void *bdrv_get_attached_dev(BlockDriverState *bs)
|
1116 |
{ |
1117 |
return bs->dev;
|
1118 |
} |
1119 |
|
1120 |
void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops, |
1121 |
void *opaque)
|
1122 |
{ |
1123 |
bs->dev_ops = ops; |
1124 |
bs->dev_opaque = opaque; |
1125 |
if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
|
1126 |
bs_snapshots = NULL;
|
1127 |
} |
1128 |
} |
1129 |
|
1130 |
void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv, |
1131 |
BlockQMPEventAction action, int is_read)
|
1132 |
{ |
1133 |
QObject *data; |
1134 |
const char *action_str; |
1135 |
|
1136 |
switch (action) {
|
1137 |
case BDRV_ACTION_REPORT:
|
1138 |
action_str = "report";
|
1139 |
break;
|
1140 |
case BDRV_ACTION_IGNORE:
|
1141 |
action_str = "ignore";
|
1142 |
break;
|
1143 |
case BDRV_ACTION_STOP:
|
1144 |
action_str = "stop";
|
1145 |
break;
|
1146 |
default:
|
1147 |
abort(); |
1148 |
} |
1149 |
|
1150 |
data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
|
1151 |
bdrv->device_name, |
1152 |
action_str, |
1153 |
is_read ? "read" : "write"); |
1154 |
monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data); |
1155 |
|
1156 |
qobject_decref(data); |
1157 |
} |
1158 |
|
1159 |
static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected) |
1160 |
{ |
1161 |
QObject *data; |
1162 |
|
1163 |
data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
|
1164 |
bdrv_get_device_name(bs), ejected); |
1165 |
monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data); |
1166 |
|
1167 |
qobject_decref(data); |
1168 |
} |
1169 |
|
1170 |
static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load) |
1171 |
{ |
1172 |
if (bs->dev_ops && bs->dev_ops->change_media_cb) {
|
1173 |
bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
|
1174 |
bs->dev_ops->change_media_cb(bs->dev_opaque, load); |
1175 |
if (tray_was_closed) {
|
1176 |
/* tray open */
|
1177 |
bdrv_emit_qmp_eject_event(bs, true);
|
1178 |
} |
1179 |
if (load) {
|
1180 |
/* tray close */
|
1181 |
bdrv_emit_qmp_eject_event(bs, false);
|
1182 |
} |
1183 |
} |
1184 |
} |
1185 |
|
1186 |
bool bdrv_dev_has_removable_media(BlockDriverState *bs)
|
1187 |
{ |
1188 |
return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
|
1189 |
} |
1190 |
|
1191 |
void bdrv_dev_eject_request(BlockDriverState *bs, bool force) |
1192 |
{ |
1193 |
if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
|
1194 |
bs->dev_ops->eject_request_cb(bs->dev_opaque, force); |
1195 |
} |
1196 |
} |
1197 |
|
1198 |
bool bdrv_dev_is_tray_open(BlockDriverState *bs)
|
1199 |
{ |
1200 |
if (bs->dev_ops && bs->dev_ops->is_tray_open) {
|
1201 |
return bs->dev_ops->is_tray_open(bs->dev_opaque);
|
1202 |
} |
1203 |
return false; |
1204 |
} |
1205 |
|
1206 |
static void bdrv_dev_resize_cb(BlockDriverState *bs) |
1207 |
{ |
1208 |
if (bs->dev_ops && bs->dev_ops->resize_cb) {
|
1209 |
bs->dev_ops->resize_cb(bs->dev_opaque); |
1210 |
} |
1211 |
} |
1212 |
|
1213 |
bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
|
1214 |
{ |
1215 |
if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
|
1216 |
return bs->dev_ops->is_medium_locked(bs->dev_opaque);
|
1217 |
} |
1218 |
return false; |
1219 |
} |
1220 |
|
1221 |
/*
|
1222 |
* Run consistency checks on an image
|
1223 |
*
|
1224 |
* Returns 0 if the check could be completed (it doesn't mean that the image is
|
1225 |
* free of errors) or -errno when an internal error occurred. The results of the
|
1226 |
* check are stored in res.
|
1227 |
*/
|
1228 |
int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
|
1229 |
{ |
1230 |
if (bs->drv->bdrv_check == NULL) { |
1231 |
return -ENOTSUP;
|
1232 |
} |
1233 |
|
1234 |
memset(res, 0, sizeof(*res)); |
1235 |
return bs->drv->bdrv_check(bs, res, fix);
|
1236 |
} |
1237 |
|
1238 |
#define COMMIT_BUF_SECTORS 2048 |
1239 |
|
1240 |
/* commit COW file into the raw image */
|
1241 |
int bdrv_commit(BlockDriverState *bs)
|
1242 |
{ |
1243 |
BlockDriver *drv = bs->drv; |
1244 |
BlockDriver *backing_drv; |
1245 |
int64_t sector, total_sectors; |
1246 |
int n, ro, open_flags;
|
1247 |
int ret = 0, rw_ret = 0; |
1248 |
uint8_t *buf; |
1249 |
char filename[1024]; |
1250 |
BlockDriverState *bs_rw, *bs_ro; |
1251 |
|
1252 |
if (!drv)
|
1253 |
return -ENOMEDIUM;
|
1254 |
|
1255 |
if (!bs->backing_hd) {
|
1256 |
return -ENOTSUP;
|
1257 |
} |
1258 |
|
1259 |
if (bs->backing_hd->keep_read_only) {
|
1260 |
return -EACCES;
|
1261 |
} |
1262 |
|
1263 |
if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
|
1264 |
return -EBUSY;
|
1265 |
} |
1266 |
|
1267 |
backing_drv = bs->backing_hd->drv; |
1268 |
ro = bs->backing_hd->read_only; |
1269 |
strncpy(filename, bs->backing_hd->filename, sizeof(filename));
|
1270 |
open_flags = bs->backing_hd->open_flags; |
1271 |
|
1272 |
if (ro) {
|
1273 |
/* re-open as RW */
|
1274 |
bdrv_delete(bs->backing_hd); |
1275 |
bs->backing_hd = NULL;
|
1276 |
bs_rw = bdrv_new("");
|
1277 |
rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, |
1278 |
backing_drv); |
1279 |
if (rw_ret < 0) { |
1280 |
bdrv_delete(bs_rw); |
1281 |
/* try to re-open read-only */
|
1282 |
bs_ro = bdrv_new("");
|
1283 |
ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, |
1284 |
backing_drv); |
1285 |
if (ret < 0) { |
1286 |
bdrv_delete(bs_ro); |
1287 |
/* drive not functional anymore */
|
1288 |
bs->drv = NULL;
|
1289 |
return ret;
|
1290 |
} |
1291 |
bs->backing_hd = bs_ro; |
1292 |
return rw_ret;
|
1293 |
} |
1294 |
bs->backing_hd = bs_rw; |
1295 |
} |
1296 |
|
1297 |
total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; |
1298 |
buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE); |
1299 |
|
1300 |
for (sector = 0; sector < total_sectors; sector += n) { |
1301 |
if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
|
1302 |
|
1303 |
if (bdrv_read(bs, sector, buf, n) != 0) { |
1304 |
ret = -EIO; |
1305 |
goto ro_cleanup;
|
1306 |
} |
1307 |
|
1308 |
if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) { |
1309 |
ret = -EIO; |
1310 |
goto ro_cleanup;
|
1311 |
} |
1312 |
} |
1313 |
} |
1314 |
|
1315 |
if (drv->bdrv_make_empty) {
|
1316 |
ret = drv->bdrv_make_empty(bs); |
1317 |
bdrv_flush(bs); |
1318 |
} |
1319 |
|
1320 |
/*
|
1321 |
* Make sure all data we wrote to the backing device is actually
|
1322 |
* stable on disk.
|
1323 |
*/
|
1324 |
if (bs->backing_hd)
|
1325 |
bdrv_flush(bs->backing_hd); |
1326 |
|
1327 |
ro_cleanup:
|
1328 |
g_free(buf); |
1329 |
|
1330 |
if (ro) {
|
1331 |
/* re-open as RO */
|
1332 |
bdrv_delete(bs->backing_hd); |
1333 |
bs->backing_hd = NULL;
|
1334 |
bs_ro = bdrv_new("");
|
1335 |
ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, |
1336 |
backing_drv); |
1337 |
if (ret < 0) { |
1338 |
bdrv_delete(bs_ro); |
1339 |
/* drive not functional anymore */
|
1340 |
bs->drv = NULL;
|
1341 |
return ret;
|
1342 |
} |
1343 |
bs->backing_hd = bs_ro; |
1344 |
bs->backing_hd->keep_read_only = 0;
|
1345 |
} |
1346 |
|
1347 |
return ret;
|
1348 |
} |
1349 |
|
1350 |
int bdrv_commit_all(void) |
1351 |
{ |
1352 |
BlockDriverState *bs; |
1353 |
|
1354 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
1355 |
int ret = bdrv_commit(bs);
|
1356 |
if (ret < 0) { |
1357 |
return ret;
|
1358 |
} |
1359 |
} |
1360 |
return 0; |
1361 |
} |
1362 |
|
1363 |
struct BdrvTrackedRequest {
|
1364 |
BlockDriverState *bs; |
1365 |
int64_t sector_num; |
1366 |
int nb_sectors;
|
1367 |
bool is_write;
|
1368 |
QLIST_ENTRY(BdrvTrackedRequest) list; |
1369 |
Coroutine *co; /* owner, used for deadlock detection */
|
1370 |
CoQueue wait_queue; /* coroutines blocked on this request */
|
1371 |
}; |
1372 |
|
1373 |
/**
|
1374 |
* Remove an active request from the tracked requests list
|
1375 |
*
|
1376 |
* This function should be called when a tracked request is completing.
|
1377 |
*/
|
1378 |
static void tracked_request_end(BdrvTrackedRequest *req) |
1379 |
{ |
1380 |
QLIST_REMOVE(req, list); |
1381 |
qemu_co_queue_restart_all(&req->wait_queue); |
1382 |
} |
1383 |
|
1384 |
/**
|
1385 |
* Add an active request to the tracked requests list
|
1386 |
*/
|
1387 |
static void tracked_request_begin(BdrvTrackedRequest *req, |
1388 |
BlockDriverState *bs, |
1389 |
int64_t sector_num, |
1390 |
int nb_sectors, bool is_write) |
1391 |
{ |
1392 |
*req = (BdrvTrackedRequest){ |
1393 |
.bs = bs, |
1394 |
.sector_num = sector_num, |
1395 |
.nb_sectors = nb_sectors, |
1396 |
.is_write = is_write, |
1397 |
.co = qemu_coroutine_self(), |
1398 |
}; |
1399 |
|
1400 |
qemu_co_queue_init(&req->wait_queue); |
1401 |
|
1402 |
QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
1403 |
} |
1404 |
|
1405 |
/**
|
1406 |
* Round a region to cluster boundaries
|
1407 |
*/
|
1408 |
static void round_to_clusters(BlockDriverState *bs, |
1409 |
int64_t sector_num, int nb_sectors,
|
1410 |
int64_t *cluster_sector_num, |
1411 |
int *cluster_nb_sectors)
|
1412 |
{ |
1413 |
BlockDriverInfo bdi; |
1414 |
|
1415 |
if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { |
1416 |
*cluster_sector_num = sector_num; |
1417 |
*cluster_nb_sectors = nb_sectors; |
1418 |
} else {
|
1419 |
int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; |
1420 |
*cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); |
1421 |
*cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + |
1422 |
nb_sectors, c); |
1423 |
} |
1424 |
} |
1425 |
|
1426 |
static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
1427 |
int64_t sector_num, int nb_sectors) {
|
1428 |
/* aaaa bbbb */
|
1429 |
if (sector_num >= req->sector_num + req->nb_sectors) {
|
1430 |
return false; |
1431 |
} |
1432 |
/* bbbb aaaa */
|
1433 |
if (req->sector_num >= sector_num + nb_sectors) {
|
1434 |
return false; |
1435 |
} |
1436 |
return true; |
1437 |
} |
1438 |
|
1439 |
static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs, |
1440 |
int64_t sector_num, int nb_sectors)
|
1441 |
{ |
1442 |
BdrvTrackedRequest *req; |
1443 |
int64_t cluster_sector_num; |
1444 |
int cluster_nb_sectors;
|
1445 |
bool retry;
|
1446 |
|
1447 |
/* If we touch the same cluster it counts as an overlap. This guarantees
|
1448 |
* that allocating writes will be serialized and not race with each other
|
1449 |
* for the same cluster. For example, in copy-on-read it ensures that the
|
1450 |
* CoR read and write operations are atomic and guest writes cannot
|
1451 |
* interleave between them.
|
1452 |
*/
|
1453 |
round_to_clusters(bs, sector_num, nb_sectors, |
1454 |
&cluster_sector_num, &cluster_nb_sectors); |
1455 |
|
1456 |
do {
|
1457 |
retry = false;
|
1458 |
QLIST_FOREACH(req, &bs->tracked_requests, list) { |
1459 |
if (tracked_request_overlaps(req, cluster_sector_num,
|
1460 |
cluster_nb_sectors)) { |
1461 |
/* Hitting this means there was a reentrant request, for
|
1462 |
* example, a block driver issuing nested requests. This must
|
1463 |
* never happen since it means deadlock.
|
1464 |
*/
|
1465 |
assert(qemu_coroutine_self() != req->co); |
1466 |
|
1467 |
qemu_co_queue_wait(&req->wait_queue); |
1468 |
retry = true;
|
1469 |
break;
|
1470 |
} |
1471 |
} |
1472 |
} while (retry);
|
1473 |
} |
1474 |
|
1475 |
/*
|
1476 |
* Return values:
|
1477 |
* 0 - success
|
1478 |
* -EINVAL - backing format specified, but no file
|
1479 |
* -ENOSPC - can't update the backing file because no space is left in the
|
1480 |
* image file header
|
1481 |
* -ENOTSUP - format driver doesn't support changing the backing file
|
1482 |
*/
|
1483 |
int bdrv_change_backing_file(BlockDriverState *bs,
|
1484 |
const char *backing_file, const char *backing_fmt) |
1485 |
{ |
1486 |
BlockDriver *drv = bs->drv; |
1487 |
int ret;
|
1488 |
|
1489 |
/* Backing file format doesn't make sense without a backing file */
|
1490 |
if (backing_fmt && !backing_file) {
|
1491 |
return -EINVAL;
|
1492 |
} |
1493 |
|
1494 |
if (drv->bdrv_change_backing_file != NULL) { |
1495 |
ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
1496 |
} else {
|
1497 |
ret = -ENOTSUP; |
1498 |
} |
1499 |
|
1500 |
if (ret == 0) { |
1501 |
pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); |
1502 |
pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); |
1503 |
} |
1504 |
return ret;
|
1505 |
} |
1506 |
|
1507 |
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, |
1508 |
size_t size) |
1509 |
{ |
1510 |
int64_t len; |
1511 |
|
1512 |
if (!bdrv_is_inserted(bs))
|
1513 |
return -ENOMEDIUM;
|
1514 |
|
1515 |
if (bs->growable)
|
1516 |
return 0; |
1517 |
|
1518 |
len = bdrv_getlength(bs); |
1519 |
|
1520 |
if (offset < 0) |
1521 |
return -EIO;
|
1522 |
|
1523 |
if ((offset > len) || (len - offset < size))
|
1524 |
return -EIO;
|
1525 |
|
1526 |
return 0; |
1527 |
} |
1528 |
|
1529 |
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, |
1530 |
int nb_sectors)
|
1531 |
{ |
1532 |
return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
|
1533 |
nb_sectors * BDRV_SECTOR_SIZE); |
1534 |
} |
1535 |
|
1536 |
typedef struct RwCo { |
1537 |
BlockDriverState *bs; |
1538 |
int64_t sector_num; |
1539 |
int nb_sectors;
|
1540 |
QEMUIOVector *qiov; |
1541 |
bool is_write;
|
1542 |
int ret;
|
1543 |
} RwCo; |
1544 |
|
1545 |
static void coroutine_fn bdrv_rw_co_entry(void *opaque) |
1546 |
{ |
1547 |
RwCo *rwco = opaque; |
1548 |
|
1549 |
if (!rwco->is_write) {
|
1550 |
rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num, |
1551 |
rwco->nb_sectors, rwco->qiov, 0);
|
1552 |
} else {
|
1553 |
rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num, |
1554 |
rwco->nb_sectors, rwco->qiov, 0);
|
1555 |
} |
1556 |
} |
1557 |
|
1558 |
/*
|
1559 |
* Process a synchronous request using coroutines
|
1560 |
*/
|
1561 |
static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, |
1562 |
int nb_sectors, bool is_write) |
1563 |
{ |
1564 |
QEMUIOVector qiov; |
1565 |
struct iovec iov = {
|
1566 |
.iov_base = (void *)buf,
|
1567 |
.iov_len = nb_sectors * BDRV_SECTOR_SIZE, |
1568 |
}; |
1569 |
Coroutine *co; |
1570 |
RwCo rwco = { |
1571 |
.bs = bs, |
1572 |
.sector_num = sector_num, |
1573 |
.nb_sectors = nb_sectors, |
1574 |
.qiov = &qiov, |
1575 |
.is_write = is_write, |
1576 |
.ret = NOT_DONE, |
1577 |
}; |
1578 |
|
1579 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
1580 |
|
1581 |
/**
|
1582 |
* In sync call context, when the vcpu is blocked, this throttling timer
|
1583 |
* will not fire; so the I/O throttling function has to be disabled here
|
1584 |
* if it has been enabled.
|
1585 |
*/
|
1586 |
if (bs->io_limits_enabled) {
|
1587 |
fprintf(stderr, "Disabling I/O throttling on '%s' due "
|
1588 |
"to synchronous I/O.\n", bdrv_get_device_name(bs));
|
1589 |
bdrv_io_limits_disable(bs); |
1590 |
} |
1591 |
|
1592 |
if (qemu_in_coroutine()) {
|
1593 |
/* Fast-path if already in coroutine context */
|
1594 |
bdrv_rw_co_entry(&rwco); |
1595 |
} else {
|
1596 |
co = qemu_coroutine_create(bdrv_rw_co_entry); |
1597 |
qemu_coroutine_enter(co, &rwco); |
1598 |
while (rwco.ret == NOT_DONE) {
|
1599 |
qemu_aio_wait(); |
1600 |
} |
1601 |
} |
1602 |
return rwco.ret;
|
1603 |
} |
1604 |
|
1605 |
/* return < 0 if error. See bdrv_write() for the return codes */
|
1606 |
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
|
1607 |
uint8_t *buf, int nb_sectors)
|
1608 |
{ |
1609 |
return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false); |
1610 |
} |
1611 |
|
1612 |
#define BITS_PER_LONG (sizeof(unsigned long) * 8) |
1613 |
|
1614 |
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num, |
1615 |
int nb_sectors, int dirty) |
1616 |
{ |
1617 |
int64_t start, end; |
1618 |
unsigned long val, idx, bit; |
1619 |
|
1620 |
start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; |
1621 |
end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
|
1622 |
|
1623 |
for (; start <= end; start++) {
|
1624 |
idx = start / BITS_PER_LONG; |
1625 |
bit = start % BITS_PER_LONG; |
1626 |
val = bs->dirty_bitmap[idx]; |
1627 |
if (dirty) {
|
1628 |
if (!(val & (1UL << bit))) { |
1629 |
bs->dirty_count++; |
1630 |
val |= 1UL << bit;
|
1631 |
} |
1632 |
} else {
|
1633 |
if (val & (1UL << bit)) { |
1634 |
bs->dirty_count--; |
1635 |
val &= ~(1UL << bit);
|
1636 |
} |
1637 |
} |
1638 |
bs->dirty_bitmap[idx] = val; |
1639 |
} |
1640 |
} |
1641 |
|
1642 |
/* Return < 0 if error. Important errors are:
|
1643 |
-EIO generic I/O error (may happen for all errors)
|
1644 |
-ENOMEDIUM No media inserted.
|
1645 |
-EINVAL Invalid sector number or nb_sectors
|
1646 |
-EACCES Trying to write a read-only device
|
1647 |
*/
|
1648 |
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
|
1649 |
const uint8_t *buf, int nb_sectors) |
1650 |
{ |
1651 |
return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true); |
1652 |
} |
1653 |
|
1654 |
int bdrv_pread(BlockDriverState *bs, int64_t offset,
|
1655 |
void *buf, int count1) |
1656 |
{ |
1657 |
uint8_t tmp_buf[BDRV_SECTOR_SIZE]; |
1658 |
int len, nb_sectors, count;
|
1659 |
int64_t sector_num; |
1660 |
int ret;
|
1661 |
|
1662 |
count = count1; |
1663 |
/* first read to align to sector start */
|
1664 |
len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
|
1665 |
if (len > count)
|
1666 |
len = count; |
1667 |
sector_num = offset >> BDRV_SECTOR_BITS; |
1668 |
if (len > 0) { |
1669 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1670 |
return ret;
|
1671 |
memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
|
1672 |
count -= len; |
1673 |
if (count == 0) |
1674 |
return count1;
|
1675 |
sector_num++; |
1676 |
buf += len; |
1677 |
} |
1678 |
|
1679 |
/* read the sectors "in place" */
|
1680 |
nb_sectors = count >> BDRV_SECTOR_BITS; |
1681 |
if (nb_sectors > 0) { |
1682 |
if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0) |
1683 |
return ret;
|
1684 |
sector_num += nb_sectors; |
1685 |
len = nb_sectors << BDRV_SECTOR_BITS; |
1686 |
buf += len; |
1687 |
count -= len; |
1688 |
} |
1689 |
|
1690 |
/* add data from the last sector */
|
1691 |
if (count > 0) { |
1692 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1693 |
return ret;
|
1694 |
memcpy(buf, tmp_buf, count); |
1695 |
} |
1696 |
return count1;
|
1697 |
} |
1698 |
|
1699 |
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
|
1700 |
const void *buf, int count1) |
1701 |
{ |
1702 |
uint8_t tmp_buf[BDRV_SECTOR_SIZE]; |
1703 |
int len, nb_sectors, count;
|
1704 |
int64_t sector_num; |
1705 |
int ret;
|
1706 |
|
1707 |
count = count1; |
1708 |
/* first write to align to sector start */
|
1709 |
len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
|
1710 |
if (len > count)
|
1711 |
len = count; |
1712 |
sector_num = offset >> BDRV_SECTOR_BITS; |
1713 |
if (len > 0) { |
1714 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1715 |
return ret;
|
1716 |
memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
|
1717 |
if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) |
1718 |
return ret;
|
1719 |
count -= len; |
1720 |
if (count == 0) |
1721 |
return count1;
|
1722 |
sector_num++; |
1723 |
buf += len; |
1724 |
} |
1725 |
|
1726 |
/* write the sectors "in place" */
|
1727 |
nb_sectors = count >> BDRV_SECTOR_BITS; |
1728 |
if (nb_sectors > 0) { |
1729 |
if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0) |
1730 |
return ret;
|
1731 |
sector_num += nb_sectors; |
1732 |
len = nb_sectors << BDRV_SECTOR_BITS; |
1733 |
buf += len; |
1734 |
count -= len; |
1735 |
} |
1736 |
|
1737 |
/* add data from the last sector */
|
1738 |
if (count > 0) { |
1739 |
if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) |
1740 |
return ret;
|
1741 |
memcpy(tmp_buf, buf, count); |
1742 |
if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) |
1743 |
return ret;
|
1744 |
} |
1745 |
return count1;
|
1746 |
} |
1747 |
|
1748 |
/*
|
1749 |
* Writes to the file and ensures that no writes are reordered across this
|
1750 |
* request (acts as a barrier)
|
1751 |
*
|
1752 |
* Returns 0 on success, -errno in error cases.
|
1753 |
*/
|
1754 |
int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
|
1755 |
const void *buf, int count) |
1756 |
{ |
1757 |
int ret;
|
1758 |
|
1759 |
ret = bdrv_pwrite(bs, offset, buf, count); |
1760 |
if (ret < 0) { |
1761 |
return ret;
|
1762 |
} |
1763 |
|
1764 |
/* No flush needed for cache modes that already do it */
|
1765 |
if (bs->enable_write_cache) {
|
1766 |
bdrv_flush(bs); |
1767 |
} |
1768 |
|
1769 |
return 0; |
1770 |
} |
1771 |
|
1772 |
static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, |
1773 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
|
1774 |
{ |
1775 |
/* Perform I/O through a temporary buffer so that users who scribble over
|
1776 |
* their read buffer while the operation is in progress do not end up
|
1777 |
* modifying the image file. This is critical for zero-copy guest I/O
|
1778 |
* where anything might happen inside guest memory.
|
1779 |
*/
|
1780 |
void *bounce_buffer;
|
1781 |
|
1782 |
BlockDriver *drv = bs->drv; |
1783 |
struct iovec iov;
|
1784 |
QEMUIOVector bounce_qiov; |
1785 |
int64_t cluster_sector_num; |
1786 |
int cluster_nb_sectors;
|
1787 |
size_t skip_bytes; |
1788 |
int ret;
|
1789 |
|
1790 |
/* Cover entire cluster so no additional backing file I/O is required when
|
1791 |
* allocating cluster in the image file.
|
1792 |
*/
|
1793 |
round_to_clusters(bs, sector_num, nb_sectors, |
1794 |
&cluster_sector_num, &cluster_nb_sectors); |
1795 |
|
1796 |
trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, |
1797 |
cluster_sector_num, cluster_nb_sectors); |
1798 |
|
1799 |
iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; |
1800 |
iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len); |
1801 |
qemu_iovec_init_external(&bounce_qiov, &iov, 1);
|
1802 |
|
1803 |
ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, |
1804 |
&bounce_qiov); |
1805 |
if (ret < 0) { |
1806 |
goto err;
|
1807 |
} |
1808 |
|
1809 |
if (drv->bdrv_co_write_zeroes &&
|
1810 |
buffer_is_zero(bounce_buffer, iov.iov_len)) { |
1811 |
ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, |
1812 |
cluster_nb_sectors); |
1813 |
} else {
|
1814 |
/* This does not change the data on the disk, it is not necessary
|
1815 |
* to flush even in cache=writethrough mode.
|
1816 |
*/
|
1817 |
ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, |
1818 |
&bounce_qiov); |
1819 |
} |
1820 |
|
1821 |
if (ret < 0) { |
1822 |
/* It might be okay to ignore write errors for guest requests. If this
|
1823 |
* is a deliberate copy-on-read then we don't want to ignore the error.
|
1824 |
* Simply report it in all cases.
|
1825 |
*/
|
1826 |
goto err;
|
1827 |
} |
1828 |
|
1829 |
skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; |
1830 |
qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes, |
1831 |
nb_sectors * BDRV_SECTOR_SIZE); |
1832 |
|
1833 |
err:
|
1834 |
qemu_vfree(bounce_buffer); |
1835 |
return ret;
|
1836 |
} |
1837 |
|
1838 |
/*
|
1839 |
* Handle a read request in coroutine context
|
1840 |
*/
|
1841 |
static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, |
1842 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
1843 |
BdrvRequestFlags flags) |
1844 |
{ |
1845 |
BlockDriver *drv = bs->drv; |
1846 |
BdrvTrackedRequest req; |
1847 |
int ret;
|
1848 |
|
1849 |
if (!drv) {
|
1850 |
return -ENOMEDIUM;
|
1851 |
} |
1852 |
if (bdrv_check_request(bs, sector_num, nb_sectors)) {
|
1853 |
return -EIO;
|
1854 |
} |
1855 |
|
1856 |
/* throttling disk read I/O */
|
1857 |
if (bs->io_limits_enabled) {
|
1858 |
bdrv_io_limits_intercept(bs, false, nb_sectors);
|
1859 |
} |
1860 |
|
1861 |
if (bs->copy_on_read) {
|
1862 |
flags |= BDRV_REQ_COPY_ON_READ; |
1863 |
} |
1864 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
1865 |
bs->copy_on_read_in_flight++; |
1866 |
} |
1867 |
|
1868 |
if (bs->copy_on_read_in_flight) {
|
1869 |
wait_for_overlapping_requests(bs, sector_num, nb_sectors); |
1870 |
} |
1871 |
|
1872 |
tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
|
1873 |
|
1874 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
1875 |
int pnum;
|
1876 |
|
1877 |
ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum); |
1878 |
if (ret < 0) { |
1879 |
goto out;
|
1880 |
} |
1881 |
|
1882 |
if (!ret || pnum != nb_sectors) {
|
1883 |
ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); |
1884 |
goto out;
|
1885 |
} |
1886 |
} |
1887 |
|
1888 |
ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1889 |
|
1890 |
out:
|
1891 |
tracked_request_end(&req); |
1892 |
|
1893 |
if (flags & BDRV_REQ_COPY_ON_READ) {
|
1894 |
bs->copy_on_read_in_flight--; |
1895 |
} |
1896 |
|
1897 |
return ret;
|
1898 |
} |
1899 |
|
1900 |
int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
|
1901 |
int nb_sectors, QEMUIOVector *qiov)
|
1902 |
{ |
1903 |
trace_bdrv_co_readv(bs, sector_num, nb_sectors); |
1904 |
|
1905 |
return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); |
1906 |
} |
1907 |
|
1908 |
int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
|
1909 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
|
1910 |
{ |
1911 |
trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); |
1912 |
|
1913 |
return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
|
1914 |
BDRV_REQ_COPY_ON_READ); |
1915 |
} |
1916 |
|
1917 |
static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, |
1918 |
int64_t sector_num, int nb_sectors)
|
1919 |
{ |
1920 |
BlockDriver *drv = bs->drv; |
1921 |
QEMUIOVector qiov; |
1922 |
struct iovec iov;
|
1923 |
int ret;
|
1924 |
|
1925 |
/* TODO Emulate only part of misaligned requests instead of letting block
|
1926 |
* drivers return -ENOTSUP and emulate everything */
|
1927 |
|
1928 |
/* First try the efficient write zeroes operation */
|
1929 |
if (drv->bdrv_co_write_zeroes) {
|
1930 |
ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors); |
1931 |
if (ret != -ENOTSUP) {
|
1932 |
return ret;
|
1933 |
} |
1934 |
} |
1935 |
|
1936 |
/* Fall back to bounce buffer if write zeroes is unsupported */
|
1937 |
iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE; |
1938 |
iov.iov_base = qemu_blockalign(bs, iov.iov_len); |
1939 |
memset(iov.iov_base, 0, iov.iov_len);
|
1940 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
1941 |
|
1942 |
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov); |
1943 |
|
1944 |
qemu_vfree(iov.iov_base); |
1945 |
return ret;
|
1946 |
} |
1947 |
|
1948 |
/*
|
1949 |
* Handle a write request in coroutine context
|
1950 |
*/
|
1951 |
static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, |
1952 |
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
1953 |
BdrvRequestFlags flags) |
1954 |
{ |
1955 |
BlockDriver *drv = bs->drv; |
1956 |
BdrvTrackedRequest req; |
1957 |
int ret;
|
1958 |
|
1959 |
if (!bs->drv) {
|
1960 |
return -ENOMEDIUM;
|
1961 |
} |
1962 |
if (bs->read_only) {
|
1963 |
return -EACCES;
|
1964 |
} |
1965 |
if (bdrv_check_request(bs, sector_num, nb_sectors)) {
|
1966 |
return -EIO;
|
1967 |
} |
1968 |
|
1969 |
/* throttling disk write I/O */
|
1970 |
if (bs->io_limits_enabled) {
|
1971 |
bdrv_io_limits_intercept(bs, true, nb_sectors);
|
1972 |
} |
1973 |
|
1974 |
if (bs->copy_on_read_in_flight) {
|
1975 |
wait_for_overlapping_requests(bs, sector_num, nb_sectors); |
1976 |
} |
1977 |
|
1978 |
tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
|
1979 |
|
1980 |
if (flags & BDRV_REQ_ZERO_WRITE) {
|
1981 |
ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors); |
1982 |
} else {
|
1983 |
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
1984 |
} |
1985 |
|
1986 |
if (ret == 0 && !bs->enable_write_cache) { |
1987 |
ret = bdrv_co_flush(bs); |
1988 |
} |
1989 |
|
1990 |
if (bs->dirty_bitmap) {
|
1991 |
set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
|
1992 |
} |
1993 |
|
1994 |
if (bs->wr_highest_sector < sector_num + nb_sectors - 1) { |
1995 |
bs->wr_highest_sector = sector_num + nb_sectors - 1;
|
1996 |
} |
1997 |
|
1998 |
tracked_request_end(&req); |
1999 |
|
2000 |
return ret;
|
2001 |
} |
2002 |
|
2003 |
int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
|
2004 |
int nb_sectors, QEMUIOVector *qiov)
|
2005 |
{ |
2006 |
trace_bdrv_co_writev(bs, sector_num, nb_sectors); |
2007 |
|
2008 |
return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); |
2009 |
} |
2010 |
|
2011 |
int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
|
2012 |
int64_t sector_num, int nb_sectors)
|
2013 |
{ |
2014 |
trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors); |
2015 |
|
2016 |
return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, |
2017 |
BDRV_REQ_ZERO_WRITE); |
2018 |
} |
2019 |
|
2020 |
/**
|
2021 |
* Truncate file to 'offset' bytes (needed only for file protocols)
|
2022 |
*/
|
2023 |
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
|
2024 |
{ |
2025 |
BlockDriver *drv = bs->drv; |
2026 |
int ret;
|
2027 |
if (!drv)
|
2028 |
return -ENOMEDIUM;
|
2029 |
if (!drv->bdrv_truncate)
|
2030 |
return -ENOTSUP;
|
2031 |
if (bs->read_only)
|
2032 |
return -EACCES;
|
2033 |
if (bdrv_in_use(bs))
|
2034 |
return -EBUSY;
|
2035 |
ret = drv->bdrv_truncate(bs, offset); |
2036 |
if (ret == 0) { |
2037 |
ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
2038 |
bdrv_dev_resize_cb(bs); |
2039 |
} |
2040 |
return ret;
|
2041 |
} |
2042 |
|
2043 |
/**
|
2044 |
* Length of a allocated file in bytes. Sparse files are counted by actual
|
2045 |
* allocated space. Return < 0 if error or unknown.
|
2046 |
*/
|
2047 |
int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) |
2048 |
{ |
2049 |
BlockDriver *drv = bs->drv; |
2050 |
if (!drv) {
|
2051 |
return -ENOMEDIUM;
|
2052 |
} |
2053 |
if (drv->bdrv_get_allocated_file_size) {
|
2054 |
return drv->bdrv_get_allocated_file_size(bs);
|
2055 |
} |
2056 |
if (bs->file) {
|
2057 |
return bdrv_get_allocated_file_size(bs->file);
|
2058 |
} |
2059 |
return -ENOTSUP;
|
2060 |
} |
2061 |
|
2062 |
/**
|
2063 |
* Length of a file in bytes. Return < 0 if error or unknown.
|
2064 |
*/
|
2065 |
int64_t bdrv_getlength(BlockDriverState *bs) |
2066 |
{ |
2067 |
BlockDriver *drv = bs->drv; |
2068 |
if (!drv)
|
2069 |
return -ENOMEDIUM;
|
2070 |
|
2071 |
if (bs->growable || bdrv_dev_has_removable_media(bs)) {
|
2072 |
if (drv->bdrv_getlength) {
|
2073 |
return drv->bdrv_getlength(bs);
|
2074 |
} |
2075 |
} |
2076 |
return bs->total_sectors * BDRV_SECTOR_SIZE;
|
2077 |
} |
2078 |
|
2079 |
/* return 0 as number of sectors if no device present or error */
|
2080 |
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
|
2081 |
{ |
2082 |
int64_t length; |
2083 |
length = bdrv_getlength(bs); |
2084 |
if (length < 0) |
2085 |
length = 0;
|
2086 |
else
|
2087 |
length = length >> BDRV_SECTOR_BITS; |
2088 |
*nb_sectors_ptr = length; |
2089 |
} |
2090 |
|
2091 |
struct partition {
|
2092 |
uint8_t boot_ind; /* 0x80 - active */
|
2093 |
uint8_t head; /* starting head */
|
2094 |
uint8_t sector; /* starting sector */
|
2095 |
uint8_t cyl; /* starting cylinder */
|
2096 |
uint8_t sys_ind; /* What partition type */
|
2097 |
uint8_t end_head; /* end head */
|
2098 |
uint8_t end_sector; /* end sector */
|
2099 |
uint8_t end_cyl; /* end cylinder */
|
2100 |
uint32_t start_sect; /* starting sector counting from 0 */
|
2101 |
uint32_t nr_sects; /* nr of sectors in partition */
|
2102 |
} QEMU_PACKED; |
2103 |
|
2104 |
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
|
2105 |
static int guess_disk_lchs(BlockDriverState *bs, |
2106 |
int *pcylinders, int *pheads, int *psectors) |
2107 |
{ |
2108 |
uint8_t buf[BDRV_SECTOR_SIZE]; |
2109 |
int ret, i, heads, sectors, cylinders;
|
2110 |
struct partition *p;
|
2111 |
uint32_t nr_sects; |
2112 |
uint64_t nb_sectors; |
2113 |
bool enabled;
|
2114 |
|
2115 |
bdrv_get_geometry(bs, &nb_sectors); |
2116 |
|
2117 |
/**
|
2118 |
* The function will be invoked during startup not only in sync I/O mode,
|
2119 |
* but also in async I/O mode. So the I/O throttling function has to
|
2120 |
* be disabled temporarily here, not permanently.
|
2121 |
*/
|
2122 |
enabled = bs->io_limits_enabled; |
2123 |
bs->io_limits_enabled = false;
|
2124 |
ret = bdrv_read(bs, 0, buf, 1); |
2125 |
bs->io_limits_enabled = enabled; |
2126 |
if (ret < 0) |
2127 |
return -1; |
2128 |
/* test msdos magic */
|
2129 |
if (buf[510] != 0x55 || buf[511] != 0xaa) |
2130 |
return -1; |
2131 |
for(i = 0; i < 4; i++) { |
2132 |
p = ((struct partition *)(buf + 0x1be)) + i; |
2133 |
nr_sects = le32_to_cpu(p->nr_sects); |
2134 |
if (nr_sects && p->end_head) {
|
2135 |
/* We make the assumption that the partition terminates on
|
2136 |
a cylinder boundary */
|
2137 |
heads = p->end_head + 1;
|
2138 |
sectors = p->end_sector & 63;
|
2139 |
if (sectors == 0) |
2140 |
continue;
|
2141 |
cylinders = nb_sectors / (heads * sectors); |
2142 |
if (cylinders < 1 || cylinders > 16383) |
2143 |
continue;
|
2144 |
*pheads = heads; |
2145 |
*psectors = sectors; |
2146 |
*pcylinders = cylinders; |
2147 |
#if 0
|
2148 |
printf("guessed geometry: LCHS=%d %d %d\n",
|
2149 |
cylinders, heads, sectors);
|
2150 |
#endif
|
2151 |
return 0; |
2152 |
} |
2153 |
} |
2154 |
return -1; |
2155 |
} |
2156 |
|
2157 |
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs) |
2158 |
{ |
2159 |
int translation, lba_detected = 0; |
2160 |
int cylinders, heads, secs;
|
2161 |
uint64_t nb_sectors; |
2162 |
|
2163 |
/* if a geometry hint is available, use it */
|
2164 |
bdrv_get_geometry(bs, &nb_sectors); |
2165 |
bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs); |
2166 |
translation = bdrv_get_translation_hint(bs); |
2167 |
if (cylinders != 0) { |
2168 |
*pcyls = cylinders; |
2169 |
*pheads = heads; |
2170 |
*psecs = secs; |
2171 |
} else {
|
2172 |
if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) { |
2173 |
if (heads > 16) { |
2174 |
/* if heads > 16, it means that a BIOS LBA
|
2175 |
translation was active, so the default
|
2176 |
hardware geometry is OK */
|
2177 |
lba_detected = 1;
|
2178 |
goto default_geometry;
|
2179 |
} else {
|
2180 |
*pcyls = cylinders; |
2181 |
*pheads = heads; |
2182 |
*psecs = secs; |
2183 |
/* disable any translation to be in sync with
|
2184 |
the logical geometry */
|
2185 |
if (translation == BIOS_ATA_TRANSLATION_AUTO) {
|
2186 |
bdrv_set_translation_hint(bs, |
2187 |
BIOS_ATA_TRANSLATION_NONE); |
2188 |
} |
2189 |
} |
2190 |
} else {
|
2191 |
default_geometry:
|
2192 |
/* if no geometry, use a standard physical disk geometry */
|
2193 |
cylinders = nb_sectors / (16 * 63); |
2194 |
|
2195 |
if (cylinders > 16383) |
2196 |
cylinders = 16383;
|
2197 |
else if (cylinders < 2) |
2198 |
cylinders = 2;
|
2199 |
*pcyls = cylinders; |
2200 |
*pheads = 16;
|
2201 |
*psecs = 63;
|
2202 |
if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) { |
2203 |
if ((*pcyls * *pheads) <= 131072) { |
2204 |
bdrv_set_translation_hint(bs, |
2205 |
BIOS_ATA_TRANSLATION_LARGE); |
2206 |
} else {
|
2207 |
bdrv_set_translation_hint(bs, |
2208 |
BIOS_ATA_TRANSLATION_LBA); |
2209 |
} |
2210 |
} |
2211 |
} |
2212 |
bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs); |
2213 |
} |
2214 |
} |
2215 |
|
2216 |
void bdrv_set_geometry_hint(BlockDriverState *bs,
|
2217 |
int cyls, int heads, int secs) |
2218 |
{ |
2219 |
bs->cyls = cyls; |
2220 |
bs->heads = heads; |
2221 |
bs->secs = secs; |
2222 |
} |
2223 |
|
2224 |
void bdrv_set_translation_hint(BlockDriverState *bs, int translation) |
2225 |
{ |
2226 |
bs->translation = translation; |
2227 |
} |
2228 |
|
2229 |
void bdrv_get_geometry_hint(BlockDriverState *bs,
|
2230 |
int *pcyls, int *pheads, int *psecs) |
2231 |
{ |
2232 |
*pcyls = bs->cyls; |
2233 |
*pheads = bs->heads; |
2234 |
*psecs = bs->secs; |
2235 |
} |
2236 |
|
2237 |
/* throttling disk io limits */
|
2238 |
void bdrv_set_io_limits(BlockDriverState *bs,
|
2239 |
BlockIOLimit *io_limits) |
2240 |
{ |
2241 |
bs->io_limits = *io_limits; |
2242 |
bs->io_limits_enabled = bdrv_io_limits_enabled(bs); |
2243 |
} |
2244 |
|
2245 |
/* Recognize floppy formats */
|
2246 |
typedef struct FDFormat { |
2247 |
FDriveType drive; |
2248 |
uint8_t last_sect; |
2249 |
uint8_t max_track; |
2250 |
uint8_t max_head; |
2251 |
FDriveRate rate; |
2252 |
} FDFormat; |
2253 |
|
2254 |
static const FDFormat fd_formats[] = { |
2255 |
/* First entry is default format */
|
2256 |
/* 1.44 MB 3"1/2 floppy disks */
|
2257 |
{ FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, }, |
2258 |
{ FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, }, |
2259 |
{ FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, }, |
2260 |
{ FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, }, |
2261 |
{ FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, }, |
2262 |
{ FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, }, |
2263 |
{ FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, }, |
2264 |
{ FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, }, |
2265 |
/* 2.88 MB 3"1/2 floppy disks */
|
2266 |
{ FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, }, |
2267 |
{ FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, }, |
2268 |
{ FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, }, |
2269 |
{ FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, }, |
2270 |
{ FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, }, |
2271 |
/* 720 kB 3"1/2 floppy disks */
|
2272 |
{ FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, }, |
2273 |
{ FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, }, |
2274 |
{ FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, }, |
2275 |
{ FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, }, |
2276 |
{ FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, }, |
2277 |
{ FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, }, |
2278 |
/* 1.2 MB 5"1/4 floppy disks */
|
2279 |
{ FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, }, |
2280 |
{ FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, }, |
2281 |
{ FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, }, |
2282 |
{ FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, }, |
2283 |
{ FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, }, |
2284 |
/* 720 kB 5"1/4 floppy disks */
|
2285 |
{ FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, }, |
2286 |
{ FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, }, |
2287 |
/* 360 kB 5"1/4 floppy disks */
|
2288 |
{ FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, }, |
2289 |
{ FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, }, |
2290 |
{ FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, }, |
2291 |
{ FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, }, |
2292 |
/* 320 kB 5"1/4 floppy disks */
|
2293 |
{ FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, }, |
2294 |
{ FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, }, |
2295 |
/* 360 kB must match 5"1/4 better than 3"1/2... */
|
2296 |
{ FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, }, |
2297 |
/* end */
|
2298 |
{ FDRIVE_DRV_NONE, -1, -1, 0, 0, }, |
2299 |
}; |
2300 |
|
2301 |
void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads, |
2302 |
int *max_track, int *last_sect, |
2303 |
FDriveType drive_in, FDriveType *drive, |
2304 |
FDriveRate *rate) |
2305 |
{ |
2306 |
const FDFormat *parse;
|
2307 |
uint64_t nb_sectors, size; |
2308 |
int i, first_match, match;
|
2309 |
|
2310 |
bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect); |
2311 |
if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) { |
2312 |
/* User defined disk */
|
2313 |
*rate = FDRIVE_RATE_500K; |
2314 |
} else {
|
2315 |
bdrv_get_geometry(bs, &nb_sectors); |
2316 |
match = -1;
|
2317 |
first_match = -1;
|
2318 |
for (i = 0; ; i++) { |
2319 |
parse = &fd_formats[i]; |
2320 |
if (parse->drive == FDRIVE_DRV_NONE) {
|
2321 |
break;
|
2322 |
} |
2323 |
if (drive_in == parse->drive ||
|
2324 |
drive_in == FDRIVE_DRV_NONE) { |
2325 |
size = (parse->max_head + 1) * parse->max_track *
|
2326 |
parse->last_sect; |
2327 |
if (nb_sectors == size) {
|
2328 |
match = i; |
2329 |
break;
|
2330 |
} |
2331 |
if (first_match == -1) { |
2332 |
first_match = i; |
2333 |
} |
2334 |
} |
2335 |
} |
2336 |
if (match == -1) { |
2337 |
if (first_match == -1) { |
2338 |
match = 1;
|
2339 |
} else {
|
2340 |
match = first_match; |
2341 |
} |
2342 |
parse = &fd_formats[match]; |
2343 |
} |
2344 |
*nb_heads = parse->max_head + 1;
|
2345 |
*max_track = parse->max_track; |
2346 |
*last_sect = parse->last_sect; |
2347 |
*drive = parse->drive; |
2348 |
*rate = parse->rate; |
2349 |
} |
2350 |
} |
2351 |
|
2352 |
int bdrv_get_translation_hint(BlockDriverState *bs)
|
2353 |
{ |
2354 |
return bs->translation;
|
2355 |
} |
2356 |
|
2357 |
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
|
2358 |
BlockErrorAction on_write_error) |
2359 |
{ |
2360 |
bs->on_read_error = on_read_error; |
2361 |
bs->on_write_error = on_write_error; |
2362 |
} |
2363 |
|
2364 |
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
|
2365 |
{ |
2366 |
return is_read ? bs->on_read_error : bs->on_write_error;
|
2367 |
} |
2368 |
|
2369 |
int bdrv_is_read_only(BlockDriverState *bs)
|
2370 |
{ |
2371 |
return bs->read_only;
|
2372 |
} |
2373 |
|
2374 |
int bdrv_is_sg(BlockDriverState *bs)
|
2375 |
{ |
2376 |
return bs->sg;
|
2377 |
} |
2378 |
|
2379 |
int bdrv_enable_write_cache(BlockDriverState *bs)
|
2380 |
{ |
2381 |
return bs->enable_write_cache;
|
2382 |
} |
2383 |
|
2384 |
void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce) |
2385 |
{ |
2386 |
bs->enable_write_cache = wce; |
2387 |
} |
2388 |
|
2389 |
int bdrv_is_encrypted(BlockDriverState *bs)
|
2390 |
{ |
2391 |
if (bs->backing_hd && bs->backing_hd->encrypted)
|
2392 |
return 1; |
2393 |
return bs->encrypted;
|
2394 |
} |
2395 |
|
2396 |
int bdrv_key_required(BlockDriverState *bs)
|
2397 |
{ |
2398 |
BlockDriverState *backing_hd = bs->backing_hd; |
2399 |
|
2400 |
if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
|
2401 |
return 1; |
2402 |
return (bs->encrypted && !bs->valid_key);
|
2403 |
} |
2404 |
|
2405 |
int bdrv_set_key(BlockDriverState *bs, const char *key) |
2406 |
{ |
2407 |
int ret;
|
2408 |
if (bs->backing_hd && bs->backing_hd->encrypted) {
|
2409 |
ret = bdrv_set_key(bs->backing_hd, key); |
2410 |
if (ret < 0) |
2411 |
return ret;
|
2412 |
if (!bs->encrypted)
|
2413 |
return 0; |
2414 |
} |
2415 |
if (!bs->encrypted) {
|
2416 |
return -EINVAL;
|
2417 |
} else if (!bs->drv || !bs->drv->bdrv_set_key) { |
2418 |
return -ENOMEDIUM;
|
2419 |
} |
2420 |
ret = bs->drv->bdrv_set_key(bs, key); |
2421 |
if (ret < 0) { |
2422 |
bs->valid_key = 0;
|
2423 |
} else if (!bs->valid_key) { |
2424 |
bs->valid_key = 1;
|
2425 |
/* call the change callback now, we skipped it on open */
|
2426 |
bdrv_dev_change_media_cb(bs, true);
|
2427 |
} |
2428 |
return ret;
|
2429 |
} |
2430 |
|
2431 |
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size) |
2432 |
{ |
2433 |
if (!bs->drv) {
|
2434 |
buf[0] = '\0'; |
2435 |
} else {
|
2436 |
pstrcpy(buf, buf_size, bs->drv->format_name); |
2437 |
} |
2438 |
} |
2439 |
|
2440 |
void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
2441 |
void *opaque)
|
2442 |
{ |
2443 |
BlockDriver *drv; |
2444 |
|
2445 |
QLIST_FOREACH(drv, &bdrv_drivers, list) { |
2446 |
it(opaque, drv->format_name); |
2447 |
} |
2448 |
} |
2449 |
|
2450 |
BlockDriverState *bdrv_find(const char *name) |
2451 |
{ |
2452 |
BlockDriverState *bs; |
2453 |
|
2454 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2455 |
if (!strcmp(name, bs->device_name)) {
|
2456 |
return bs;
|
2457 |
} |
2458 |
} |
2459 |
return NULL; |
2460 |
} |
2461 |
|
2462 |
BlockDriverState *bdrv_next(BlockDriverState *bs) |
2463 |
{ |
2464 |
if (!bs) {
|
2465 |
return QTAILQ_FIRST(&bdrv_states);
|
2466 |
} |
2467 |
return QTAILQ_NEXT(bs, list);
|
2468 |
} |
2469 |
|
2470 |
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque) |
2471 |
{ |
2472 |
BlockDriverState *bs; |
2473 |
|
2474 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2475 |
it(opaque, bs); |
2476 |
} |
2477 |
} |
2478 |
|
2479 |
const char *bdrv_get_device_name(BlockDriverState *bs) |
2480 |
{ |
2481 |
return bs->device_name;
|
2482 |
} |
2483 |
|
2484 |
int bdrv_get_flags(BlockDriverState *bs)
|
2485 |
{ |
2486 |
return bs->open_flags;
|
2487 |
} |
2488 |
|
2489 |
void bdrv_flush_all(void) |
2490 |
{ |
2491 |
BlockDriverState *bs; |
2492 |
|
2493 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2494 |
bdrv_flush(bs); |
2495 |
} |
2496 |
} |
2497 |
|
2498 |
int bdrv_has_zero_init(BlockDriverState *bs)
|
2499 |
{ |
2500 |
assert(bs->drv); |
2501 |
|
2502 |
if (bs->drv->bdrv_has_zero_init) {
|
2503 |
return bs->drv->bdrv_has_zero_init(bs);
|
2504 |
} |
2505 |
|
2506 |
return 1; |
2507 |
} |
2508 |
|
2509 |
typedef struct BdrvCoIsAllocatedData { |
2510 |
BlockDriverState *bs; |
2511 |
int64_t sector_num; |
2512 |
int nb_sectors;
|
2513 |
int *pnum;
|
2514 |
int ret;
|
2515 |
bool done;
|
2516 |
} BdrvCoIsAllocatedData; |
2517 |
|
2518 |
/*
|
2519 |
* Returns true iff the specified sector is present in the disk image. Drivers
|
2520 |
* not implementing the functionality are assumed to not support backing files,
|
2521 |
* hence all their sectors are reported as allocated.
|
2522 |
*
|
2523 |
* If 'sector_num' is beyond the end of the disk image the return value is 0
|
2524 |
* and 'pnum' is set to 0.
|
2525 |
*
|
2526 |
* 'pnum' is set to the number of sectors (including and immediately following
|
2527 |
* the specified sector) that are known to be in the same
|
2528 |
* allocated/unallocated state.
|
2529 |
*
|
2530 |
* 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
|
2531 |
* beyond the end of the disk image it will be clamped.
|
2532 |
*/
|
2533 |
int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
|
2534 |
int nb_sectors, int *pnum) |
2535 |
{ |
2536 |
int64_t n; |
2537 |
|
2538 |
if (sector_num >= bs->total_sectors) {
|
2539 |
*pnum = 0;
|
2540 |
return 0; |
2541 |
} |
2542 |
|
2543 |
n = bs->total_sectors - sector_num; |
2544 |
if (n < nb_sectors) {
|
2545 |
nb_sectors = n; |
2546 |
} |
2547 |
|
2548 |
if (!bs->drv->bdrv_co_is_allocated) {
|
2549 |
*pnum = nb_sectors; |
2550 |
return 1; |
2551 |
} |
2552 |
|
2553 |
return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
|
2554 |
} |
2555 |
|
2556 |
/* Coroutine wrapper for bdrv_is_allocated() */
|
2557 |
static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque) |
2558 |
{ |
2559 |
BdrvCoIsAllocatedData *data = opaque; |
2560 |
BlockDriverState *bs = data->bs; |
2561 |
|
2562 |
data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors, |
2563 |
data->pnum); |
2564 |
data->done = true;
|
2565 |
} |
2566 |
|
2567 |
/*
|
2568 |
* Synchronous wrapper around bdrv_co_is_allocated().
|
2569 |
*
|
2570 |
* See bdrv_co_is_allocated() for details.
|
2571 |
*/
|
2572 |
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors, |
2573 |
int *pnum)
|
2574 |
{ |
2575 |
Coroutine *co; |
2576 |
BdrvCoIsAllocatedData data = { |
2577 |
.bs = bs, |
2578 |
.sector_num = sector_num, |
2579 |
.nb_sectors = nb_sectors, |
2580 |
.pnum = pnum, |
2581 |
.done = false,
|
2582 |
}; |
2583 |
|
2584 |
co = qemu_coroutine_create(bdrv_is_allocated_co_entry); |
2585 |
qemu_coroutine_enter(co, &data); |
2586 |
while (!data.done) {
|
2587 |
qemu_aio_wait(); |
2588 |
} |
2589 |
return data.ret;
|
2590 |
} |
2591 |
|
2592 |
/*
|
2593 |
* Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
|
2594 |
*
|
2595 |
* Return true if the given sector is allocated in any image between
|
2596 |
* BASE and TOP (inclusive). BASE can be NULL to check if the given
|
2597 |
* sector is allocated in any image of the chain. Return false otherwise.
|
2598 |
*
|
2599 |
* 'pnum' is set to the number of sectors (including and immediately following
|
2600 |
* the specified sector) that are known to be in the same
|
2601 |
* allocated/unallocated state.
|
2602 |
*
|
2603 |
*/
|
2604 |
int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top,
|
2605 |
BlockDriverState *base, |
2606 |
int64_t sector_num, |
2607 |
int nb_sectors, int *pnum) |
2608 |
{ |
2609 |
BlockDriverState *intermediate; |
2610 |
int ret, n = nb_sectors;
|
2611 |
|
2612 |
intermediate = top; |
2613 |
while (intermediate && intermediate != base) {
|
2614 |
int pnum_inter;
|
2615 |
ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors, |
2616 |
&pnum_inter); |
2617 |
if (ret < 0) { |
2618 |
return ret;
|
2619 |
} else if (ret) { |
2620 |
*pnum = pnum_inter; |
2621 |
return 1; |
2622 |
} |
2623 |
|
2624 |
/*
|
2625 |
* [sector_num, nb_sectors] is unallocated on top but intermediate
|
2626 |
* might have
|
2627 |
*
|
2628 |
* [sector_num+x, nr_sectors] allocated.
|
2629 |
*/
|
2630 |
if (n > pnum_inter) {
|
2631 |
n = pnum_inter; |
2632 |
} |
2633 |
|
2634 |
intermediate = intermediate->backing_hd; |
2635 |
} |
2636 |
|
2637 |
*pnum = n; |
2638 |
return 0; |
2639 |
} |
2640 |
|
2641 |
BlockInfoList *qmp_query_block(Error **errp) |
2642 |
{ |
2643 |
BlockInfoList *head = NULL, *cur_item = NULL; |
2644 |
BlockDriverState *bs; |
2645 |
|
2646 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2647 |
BlockInfoList *info = g_malloc0(sizeof(*info));
|
2648 |
|
2649 |
info->value = g_malloc0(sizeof(*info->value));
|
2650 |
info->value->device = g_strdup(bs->device_name); |
2651 |
info->value->type = g_strdup("unknown");
|
2652 |
info->value->locked = bdrv_dev_is_medium_locked(bs); |
2653 |
info->value->removable = bdrv_dev_has_removable_media(bs); |
2654 |
|
2655 |
if (bdrv_dev_has_removable_media(bs)) {
|
2656 |
info->value->has_tray_open = true;
|
2657 |
info->value->tray_open = bdrv_dev_is_tray_open(bs); |
2658 |
} |
2659 |
|
2660 |
if (bdrv_iostatus_is_enabled(bs)) {
|
2661 |
info->value->has_io_status = true;
|
2662 |
info->value->io_status = bs->iostatus; |
2663 |
} |
2664 |
|
2665 |
if (bs->drv) {
|
2666 |
info->value->has_inserted = true;
|
2667 |
info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
|
2668 |
info->value->inserted->file = g_strdup(bs->filename); |
2669 |
info->value->inserted->ro = bs->read_only; |
2670 |
info->value->inserted->drv = g_strdup(bs->drv->format_name); |
2671 |
info->value->inserted->encrypted = bs->encrypted; |
2672 |
if (bs->backing_file[0]) { |
2673 |
info->value->inserted->has_backing_file = true;
|
2674 |
info->value->inserted->backing_file = g_strdup(bs->backing_file); |
2675 |
} |
2676 |
|
2677 |
if (bs->io_limits_enabled) {
|
2678 |
info->value->inserted->bps = |
2679 |
bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; |
2680 |
info->value->inserted->bps_rd = |
2681 |
bs->io_limits.bps[BLOCK_IO_LIMIT_READ]; |
2682 |
info->value->inserted->bps_wr = |
2683 |
bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE]; |
2684 |
info->value->inserted->iops = |
2685 |
bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; |
2686 |
info->value->inserted->iops_rd = |
2687 |
bs->io_limits.iops[BLOCK_IO_LIMIT_READ]; |
2688 |
info->value->inserted->iops_wr = |
2689 |
bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE]; |
2690 |
} |
2691 |
} |
2692 |
|
2693 |
/* XXX: waiting for the qapi to support GSList */
|
2694 |
if (!cur_item) {
|
2695 |
head = cur_item = info; |
2696 |
} else {
|
2697 |
cur_item->next = info; |
2698 |
cur_item = info; |
2699 |
} |
2700 |
} |
2701 |
|
2702 |
return head;
|
2703 |
} |
2704 |
|
2705 |
/* Consider exposing this as a full fledged QMP command */
|
2706 |
static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp) |
2707 |
{ |
2708 |
BlockStats *s; |
2709 |
|
2710 |
s = g_malloc0(sizeof(*s));
|
2711 |
|
2712 |
if (bs->device_name[0]) { |
2713 |
s->has_device = true;
|
2714 |
s->device = g_strdup(bs->device_name); |
2715 |
} |
2716 |
|
2717 |
s->stats = g_malloc0(sizeof(*s->stats));
|
2718 |
s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ]; |
2719 |
s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE]; |
2720 |
s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ]; |
2721 |
s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE]; |
2722 |
s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE; |
2723 |
s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH]; |
2724 |
s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE]; |
2725 |
s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ]; |
2726 |
s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH]; |
2727 |
|
2728 |
if (bs->file) {
|
2729 |
s->has_parent = true;
|
2730 |
s->parent = qmp_query_blockstat(bs->file, NULL);
|
2731 |
} |
2732 |
|
2733 |
return s;
|
2734 |
} |
2735 |
|
2736 |
BlockStatsList *qmp_query_blockstats(Error **errp) |
2737 |
{ |
2738 |
BlockStatsList *head = NULL, *cur_item = NULL; |
2739 |
BlockDriverState *bs; |
2740 |
|
2741 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
2742 |
BlockStatsList *info = g_malloc0(sizeof(*info));
|
2743 |
info->value = qmp_query_blockstat(bs, NULL);
|
2744 |
|
2745 |
/* XXX: waiting for the qapi to support GSList */
|
2746 |
if (!cur_item) {
|
2747 |
head = cur_item = info; |
2748 |
} else {
|
2749 |
cur_item->next = info; |
2750 |
cur_item = info; |
2751 |
} |
2752 |
} |
2753 |
|
2754 |
return head;
|
2755 |
} |
2756 |
|
2757 |
const char *bdrv_get_encrypted_filename(BlockDriverState *bs) |
2758 |
{ |
2759 |
if (bs->backing_hd && bs->backing_hd->encrypted)
|
2760 |
return bs->backing_file;
|
2761 |
else if (bs->encrypted) |
2762 |
return bs->filename;
|
2763 |
else
|
2764 |
return NULL; |
2765 |
} |
2766 |
|
2767 |
void bdrv_get_backing_filename(BlockDriverState *bs,
|
2768 |
char *filename, int filename_size) |
2769 |
{ |
2770 |
pstrcpy(filename, filename_size, bs->backing_file); |
2771 |
} |
2772 |
|
2773 |
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
|
2774 |
const uint8_t *buf, int nb_sectors) |
2775 |
{ |
2776 |
BlockDriver *drv = bs->drv; |
2777 |
if (!drv)
|
2778 |
return -ENOMEDIUM;
|
2779 |
if (!drv->bdrv_write_compressed)
|
2780 |
return -ENOTSUP;
|
2781 |
if (bdrv_check_request(bs, sector_num, nb_sectors))
|
2782 |
return -EIO;
|
2783 |
|
2784 |
if (bs->dirty_bitmap) {
|
2785 |
set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
|
2786 |
} |
2787 |
|
2788 |
return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
|
2789 |
} |
2790 |
|
2791 |
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
|
2792 |
{ |
2793 |
BlockDriver *drv = bs->drv; |
2794 |
if (!drv)
|
2795 |
return -ENOMEDIUM;
|
2796 |
if (!drv->bdrv_get_info)
|
2797 |
return -ENOTSUP;
|
2798 |
memset(bdi, 0, sizeof(*bdi)); |
2799 |
return drv->bdrv_get_info(bs, bdi);
|
2800 |
} |
2801 |
|
2802 |
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2803 |
int64_t pos, int size)
|
2804 |
{ |
2805 |
BlockDriver *drv = bs->drv; |
2806 |
if (!drv)
|
2807 |
return -ENOMEDIUM;
|
2808 |
if (drv->bdrv_save_vmstate)
|
2809 |
return drv->bdrv_save_vmstate(bs, buf, pos, size);
|
2810 |
if (bs->file)
|
2811 |
return bdrv_save_vmstate(bs->file, buf, pos, size);
|
2812 |
return -ENOTSUP;
|
2813 |
} |
2814 |
|
2815 |
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
|
2816 |
int64_t pos, int size)
|
2817 |
{ |
2818 |
BlockDriver *drv = bs->drv; |
2819 |
if (!drv)
|
2820 |
return -ENOMEDIUM;
|
2821 |
if (drv->bdrv_load_vmstate)
|
2822 |
return drv->bdrv_load_vmstate(bs, buf, pos, size);
|
2823 |
if (bs->file)
|
2824 |
return bdrv_load_vmstate(bs->file, buf, pos, size);
|
2825 |
return -ENOTSUP;
|
2826 |
} |
2827 |
|
2828 |
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
|
2829 |
{ |
2830 |
BlockDriver *drv = bs->drv; |
2831 |
|
2832 |
if (!drv || !drv->bdrv_debug_event) {
|
2833 |
return;
|
2834 |
} |
2835 |
|
2836 |
return drv->bdrv_debug_event(bs, event);
|
2837 |
|
2838 |
} |
2839 |
|
2840 |
/**************************************************************/
|
2841 |
/* handling of snapshots */
|
2842 |
|
2843 |
int bdrv_can_snapshot(BlockDriverState *bs)
|
2844 |
{ |
2845 |
BlockDriver *drv = bs->drv; |
2846 |
if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
|
2847 |
return 0; |
2848 |
} |
2849 |
|
2850 |
if (!drv->bdrv_snapshot_create) {
|
2851 |
if (bs->file != NULL) { |
2852 |
return bdrv_can_snapshot(bs->file);
|
2853 |
} |
2854 |
return 0; |
2855 |
} |
2856 |
|
2857 |
return 1; |
2858 |
} |
2859 |
|
2860 |
int bdrv_is_snapshot(BlockDriverState *bs)
|
2861 |
{ |
2862 |
return !!(bs->open_flags & BDRV_O_SNAPSHOT);
|
2863 |
} |
2864 |
|
2865 |
BlockDriverState *bdrv_snapshots(void)
|
2866 |
{ |
2867 |
BlockDriverState *bs; |
2868 |
|
2869 |
if (bs_snapshots) {
|
2870 |
return bs_snapshots;
|
2871 |
} |
2872 |
|
2873 |
bs = NULL;
|
2874 |
while ((bs = bdrv_next(bs))) {
|
2875 |
if (bdrv_can_snapshot(bs)) {
|
2876 |
bs_snapshots = bs; |
2877 |
return bs;
|
2878 |
} |
2879 |
} |
2880 |
return NULL; |
2881 |
} |
2882 |
|
2883 |
int bdrv_snapshot_create(BlockDriverState *bs,
|
2884 |
QEMUSnapshotInfo *sn_info) |
2885 |
{ |
2886 |
BlockDriver *drv = bs->drv; |
2887 |
if (!drv)
|
2888 |
return -ENOMEDIUM;
|
2889 |
if (drv->bdrv_snapshot_create)
|
2890 |
return drv->bdrv_snapshot_create(bs, sn_info);
|
2891 |
if (bs->file)
|
2892 |
return bdrv_snapshot_create(bs->file, sn_info);
|
2893 |
return -ENOTSUP;
|
2894 |
} |
2895 |
|
2896 |
int bdrv_snapshot_goto(BlockDriverState *bs,
|
2897 |
const char *snapshot_id) |
2898 |
{ |
2899 |
BlockDriver *drv = bs->drv; |
2900 |
int ret, open_ret;
|
2901 |
|
2902 |
if (!drv)
|
2903 |
return -ENOMEDIUM;
|
2904 |
if (drv->bdrv_snapshot_goto)
|
2905 |
return drv->bdrv_snapshot_goto(bs, snapshot_id);
|
2906 |
|
2907 |
if (bs->file) {
|
2908 |
drv->bdrv_close(bs); |
2909 |
ret = bdrv_snapshot_goto(bs->file, snapshot_id); |
2910 |
open_ret = drv->bdrv_open(bs, bs->open_flags); |
2911 |
if (open_ret < 0) { |
2912 |
bdrv_delete(bs->file); |
2913 |
bs->drv = NULL;
|
2914 |
return open_ret;
|
2915 |
} |
2916 |
return ret;
|
2917 |
} |
2918 |
|
2919 |
return -ENOTSUP;
|
2920 |
} |
2921 |
|
2922 |
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) |
2923 |
{ |
2924 |
BlockDriver *drv = bs->drv; |
2925 |
if (!drv)
|
2926 |
return -ENOMEDIUM;
|
2927 |
if (drv->bdrv_snapshot_delete)
|
2928 |
return drv->bdrv_snapshot_delete(bs, snapshot_id);
|
2929 |
if (bs->file)
|
2930 |
return bdrv_snapshot_delete(bs->file, snapshot_id);
|
2931 |
return -ENOTSUP;
|
2932 |
} |
2933 |
|
2934 |
int bdrv_snapshot_list(BlockDriverState *bs,
|
2935 |
QEMUSnapshotInfo **psn_info) |
2936 |
{ |
2937 |
BlockDriver *drv = bs->drv; |
2938 |
if (!drv)
|
2939 |
return -ENOMEDIUM;
|
2940 |
if (drv->bdrv_snapshot_list)
|
2941 |
return drv->bdrv_snapshot_list(bs, psn_info);
|
2942 |
if (bs->file)
|
2943 |
return bdrv_snapshot_list(bs->file, psn_info);
|
2944 |
return -ENOTSUP;
|
2945 |
} |
2946 |
|
2947 |
int bdrv_snapshot_load_tmp(BlockDriverState *bs,
|
2948 |
const char *snapshot_name) |
2949 |
{ |
2950 |
BlockDriver *drv = bs->drv; |
2951 |
if (!drv) {
|
2952 |
return -ENOMEDIUM;
|
2953 |
} |
2954 |
if (!bs->read_only) {
|
2955 |
return -EINVAL;
|
2956 |
} |
2957 |
if (drv->bdrv_snapshot_load_tmp) {
|
2958 |
return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
|
2959 |
} |
2960 |
return -ENOTSUP;
|
2961 |
} |
2962 |
|
2963 |
BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, |
2964 |
const char *backing_file) |
2965 |
{ |
2966 |
if (!bs->drv) {
|
2967 |
return NULL; |
2968 |
} |
2969 |
|
2970 |
if (bs->backing_hd) {
|
2971 |
if (strcmp(bs->backing_file, backing_file) == 0) { |
2972 |
return bs->backing_hd;
|
2973 |
} else {
|
2974 |
return bdrv_find_backing_image(bs->backing_hd, backing_file);
|
2975 |
} |
2976 |
} |
2977 |
|
2978 |
return NULL; |
2979 |
} |
2980 |
|
2981 |
#define NB_SUFFIXES 4 |
2982 |
|
2983 |
char *get_human_readable_size(char *buf, int buf_size, int64_t size) |
2984 |
{ |
2985 |
static const char suffixes[NB_SUFFIXES] = "KMGT"; |
2986 |
int64_t base; |
2987 |
int i;
|
2988 |
|
2989 |
if (size <= 999) { |
2990 |
snprintf(buf, buf_size, "%" PRId64, size);
|
2991 |
} else {
|
2992 |
base = 1024;
|
2993 |
for(i = 0; i < NB_SUFFIXES; i++) { |
2994 |
if (size < (10 * base)) { |
2995 |
snprintf(buf, buf_size, "%0.1f%c",
|
2996 |
(double)size / base,
|
2997 |
suffixes[i]); |
2998 |
break;
|
2999 |
} else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) { |
3000 |
snprintf(buf, buf_size, "%" PRId64 "%c", |
3001 |
((size + (base >> 1)) / base),
|
3002 |
suffixes[i]); |
3003 |
break;
|
3004 |
} |
3005 |
base = base * 1024;
|
3006 |
} |
3007 |
} |
3008 |
return buf;
|
3009 |
} |
3010 |
|
3011 |
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn) |
3012 |
{ |
3013 |
char buf1[128], date_buf[128], clock_buf[128]; |
3014 |
#ifdef _WIN32
|
3015 |
struct tm *ptm;
|
3016 |
#else
|
3017 |
struct tm tm;
|
3018 |
#endif
|
3019 |
time_t ti; |
3020 |
int64_t secs; |
3021 |
|
3022 |
if (!sn) {
|
3023 |
snprintf(buf, buf_size, |
3024 |
"%-10s%-20s%7s%20s%15s",
|
3025 |
"ID", "TAG", "VM SIZE", "DATE", "VM CLOCK"); |
3026 |
} else {
|
3027 |
ti = sn->date_sec; |
3028 |
#ifdef _WIN32
|
3029 |
ptm = localtime(&ti); |
3030 |
strftime(date_buf, sizeof(date_buf),
|
3031 |
"%Y-%m-%d %H:%M:%S", ptm);
|
3032 |
#else
|
3033 |
localtime_r(&ti, &tm); |
3034 |
strftime(date_buf, sizeof(date_buf),
|
3035 |
"%Y-%m-%d %H:%M:%S", &tm);
|
3036 |
#endif
|
3037 |
secs = sn->vm_clock_nsec / 1000000000;
|
3038 |
snprintf(clock_buf, sizeof(clock_buf),
|
3039 |
"%02d:%02d:%02d.%03d",
|
3040 |
(int)(secs / 3600), |
3041 |
(int)((secs / 60) % 60), |
3042 |
(int)(secs % 60), |
3043 |
(int)((sn->vm_clock_nsec / 1000000) % 1000)); |
3044 |
snprintf(buf, buf_size, |
3045 |
"%-10s%-20s%7s%20s%15s",
|
3046 |
sn->id_str, sn->name, |
3047 |
get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
|
3048 |
date_buf, |
3049 |
clock_buf); |
3050 |
} |
3051 |
return buf;
|
3052 |
} |
3053 |
|
3054 |
/**************************************************************/
|
3055 |
/* async I/Os */
|
3056 |
|
3057 |
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, |
3058 |
QEMUIOVector *qiov, int nb_sectors,
|
3059 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3060 |
{ |
3061 |
trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); |
3062 |
|
3063 |
return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
|
3064 |
cb, opaque, false);
|
3065 |
} |
3066 |
|
3067 |
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, |
3068 |
QEMUIOVector *qiov, int nb_sectors,
|
3069 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3070 |
{ |
3071 |
trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); |
3072 |
|
3073 |
return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
|
3074 |
cb, opaque, true);
|
3075 |
} |
3076 |
|
3077 |
|
3078 |
typedef struct MultiwriteCB { |
3079 |
int error;
|
3080 |
int num_requests;
|
3081 |
int num_callbacks;
|
3082 |
struct {
|
3083 |
BlockDriverCompletionFunc *cb; |
3084 |
void *opaque;
|
3085 |
QEMUIOVector *free_qiov; |
3086 |
} callbacks[]; |
3087 |
} MultiwriteCB; |
3088 |
|
3089 |
static void multiwrite_user_cb(MultiwriteCB *mcb) |
3090 |
{ |
3091 |
int i;
|
3092 |
|
3093 |
for (i = 0; i < mcb->num_callbacks; i++) { |
3094 |
mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); |
3095 |
if (mcb->callbacks[i].free_qiov) {
|
3096 |
qemu_iovec_destroy(mcb->callbacks[i].free_qiov); |
3097 |
} |
3098 |
g_free(mcb->callbacks[i].free_qiov); |
3099 |
} |
3100 |
} |
3101 |
|
3102 |
static void multiwrite_cb(void *opaque, int ret) |
3103 |
{ |
3104 |
MultiwriteCB *mcb = opaque; |
3105 |
|
3106 |
trace_multiwrite_cb(mcb, ret); |
3107 |
|
3108 |
if (ret < 0 && !mcb->error) { |
3109 |
mcb->error = ret; |
3110 |
} |
3111 |
|
3112 |
mcb->num_requests--; |
3113 |
if (mcb->num_requests == 0) { |
3114 |
multiwrite_user_cb(mcb); |
3115 |
g_free(mcb); |
3116 |
} |
3117 |
} |
3118 |
|
3119 |
static int multiwrite_req_compare(const void *a, const void *b) |
3120 |
{ |
3121 |
const BlockRequest *req1 = a, *req2 = b;
|
3122 |
|
3123 |
/*
|
3124 |
* Note that we can't simply subtract req2->sector from req1->sector
|
3125 |
* here as that could overflow the return value.
|
3126 |
*/
|
3127 |
if (req1->sector > req2->sector) {
|
3128 |
return 1; |
3129 |
} else if (req1->sector < req2->sector) { |
3130 |
return -1; |
3131 |
} else {
|
3132 |
return 0; |
3133 |
} |
3134 |
} |
3135 |
|
3136 |
/*
|
3137 |
* Takes a bunch of requests and tries to merge them. Returns the number of
|
3138 |
* requests that remain after merging.
|
3139 |
*/
|
3140 |
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, |
3141 |
int num_reqs, MultiwriteCB *mcb)
|
3142 |
{ |
3143 |
int i, outidx;
|
3144 |
|
3145 |
// Sort requests by start sector
|
3146 |
qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
|
3147 |
|
3148 |
// Check if adjacent requests touch the same clusters. If so, combine them,
|
3149 |
// filling up gaps with zero sectors.
|
3150 |
outidx = 0;
|
3151 |
for (i = 1; i < num_reqs; i++) { |
3152 |
int merge = 0; |
3153 |
int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; |
3154 |
|
3155 |
// Handle exactly sequential writes and overlapping writes.
|
3156 |
if (reqs[i].sector <= oldreq_last) {
|
3157 |
merge = 1;
|
3158 |
} |
3159 |
|
3160 |
if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { |
3161 |
merge = 0;
|
3162 |
} |
3163 |
|
3164 |
if (merge) {
|
3165 |
size_t size; |
3166 |
QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
|
3167 |
qemu_iovec_init(qiov, |
3168 |
reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
|
3169 |
|
3170 |
// Add the first request to the merged one. If the requests are
|
3171 |
// overlapping, drop the last sectors of the first request.
|
3172 |
size = (reqs[i].sector - reqs[outidx].sector) << 9;
|
3173 |
qemu_iovec_concat(qiov, reqs[outidx].qiov, size); |
3174 |
|
3175 |
// We should need to add any zeros between the two requests
|
3176 |
assert (reqs[i].sector <= oldreq_last); |
3177 |
|
3178 |
// Add the second request
|
3179 |
qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size); |
3180 |
|
3181 |
reqs[outidx].nb_sectors = qiov->size >> 9;
|
3182 |
reqs[outidx].qiov = qiov; |
3183 |
|
3184 |
mcb->callbacks[i].free_qiov = reqs[outidx].qiov; |
3185 |
} else {
|
3186 |
outidx++; |
3187 |
reqs[outidx].sector = reqs[i].sector; |
3188 |
reqs[outidx].nb_sectors = reqs[i].nb_sectors; |
3189 |
reqs[outidx].qiov = reqs[i].qiov; |
3190 |
} |
3191 |
} |
3192 |
|
3193 |
return outidx + 1; |
3194 |
} |
3195 |
|
3196 |
/*
|
3197 |
* Submit multiple AIO write requests at once.
|
3198 |
*
|
3199 |
* On success, the function returns 0 and all requests in the reqs array have
|
3200 |
* been submitted. In error case this function returns -1, and any of the
|
3201 |
* requests may or may not be submitted yet. In particular, this means that the
|
3202 |
* callback will be called for some of the requests, for others it won't. The
|
3203 |
* caller must check the error field of the BlockRequest to wait for the right
|
3204 |
* callbacks (if error != 0, no callback will be called).
|
3205 |
*
|
3206 |
* The implementation may modify the contents of the reqs array, e.g. to merge
|
3207 |
* requests. However, the fields opaque and error are left unmodified as they
|
3208 |
* are used to signal failure for a single request to the caller.
|
3209 |
*/
|
3210 |
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) |
3211 |
{ |
3212 |
MultiwriteCB *mcb; |
3213 |
int i;
|
3214 |
|
3215 |
/* don't submit writes if we don't have a medium */
|
3216 |
if (bs->drv == NULL) { |
3217 |
for (i = 0; i < num_reqs; i++) { |
3218 |
reqs[i].error = -ENOMEDIUM; |
3219 |
} |
3220 |
return -1; |
3221 |
} |
3222 |
|
3223 |
if (num_reqs == 0) { |
3224 |
return 0; |
3225 |
} |
3226 |
|
3227 |
// Create MultiwriteCB structure
|
3228 |
mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); |
3229 |
mcb->num_requests = 0;
|
3230 |
mcb->num_callbacks = num_reqs; |
3231 |
|
3232 |
for (i = 0; i < num_reqs; i++) { |
3233 |
mcb->callbacks[i].cb = reqs[i].cb; |
3234 |
mcb->callbacks[i].opaque = reqs[i].opaque; |
3235 |
} |
3236 |
|
3237 |
// Check for mergable requests
|
3238 |
num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); |
3239 |
|
3240 |
trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); |
3241 |
|
3242 |
/* Run the aio requests. */
|
3243 |
mcb->num_requests = num_reqs; |
3244 |
for (i = 0; i < num_reqs; i++) { |
3245 |
bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov, |
3246 |
reqs[i].nb_sectors, multiwrite_cb, mcb); |
3247 |
} |
3248 |
|
3249 |
return 0; |
3250 |
} |
3251 |
|
3252 |
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
|
3253 |
{ |
3254 |
acb->pool->cancel(acb); |
3255 |
} |
3256 |
|
3257 |
/* block I/O throttling */
|
3258 |
static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors, |
3259 |
bool is_write, double elapsed_time, uint64_t *wait) |
3260 |
{ |
3261 |
uint64_t bps_limit = 0;
|
3262 |
double bytes_limit, bytes_base, bytes_res;
|
3263 |
double slice_time, wait_time;
|
3264 |
|
3265 |
if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
|
3266 |
bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; |
3267 |
} else if (bs->io_limits.bps[is_write]) { |
3268 |
bps_limit = bs->io_limits.bps[is_write]; |
3269 |
} else {
|
3270 |
if (wait) {
|
3271 |
*wait = 0;
|
3272 |
} |
3273 |
|
3274 |
return false; |
3275 |
} |
3276 |
|
3277 |
slice_time = bs->slice_end - bs->slice_start; |
3278 |
slice_time /= (NANOSECONDS_PER_SECOND); |
3279 |
bytes_limit = bps_limit * slice_time; |
3280 |
bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write]; |
3281 |
if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
|
3282 |
bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write]; |
3283 |
} |
3284 |
|
3285 |
/* bytes_base: the bytes of data which have been read/written; and
|
3286 |
* it is obtained from the history statistic info.
|
3287 |
* bytes_res: the remaining bytes of data which need to be read/written.
|
3288 |
* (bytes_base + bytes_res) / bps_limit: used to calcuate
|
3289 |
* the total time for completing reading/writting all data.
|
3290 |
*/
|
3291 |
bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
|
3292 |
|
3293 |
if (bytes_base + bytes_res <= bytes_limit) {
|
3294 |
if (wait) {
|
3295 |
*wait = 0;
|
3296 |
} |
3297 |
|
3298 |
return false; |
3299 |
} |
3300 |
|
3301 |
/* Calc approx time to dispatch */
|
3302 |
wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time; |
3303 |
|
3304 |
/* When the I/O rate at runtime exceeds the limits,
|
3305 |
* bs->slice_end need to be extended in order that the current statistic
|
3306 |
* info can be kept until the timer fire, so it is increased and tuned
|
3307 |
* based on the result of experiment.
|
3308 |
*/
|
3309 |
bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3310 |
bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
|
3311 |
if (wait) {
|
3312 |
*wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3313 |
} |
3314 |
|
3315 |
return true; |
3316 |
} |
3317 |
|
3318 |
static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write, |
3319 |
double elapsed_time, uint64_t *wait)
|
3320 |
{ |
3321 |
uint64_t iops_limit = 0;
|
3322 |
double ios_limit, ios_base;
|
3323 |
double slice_time, wait_time;
|
3324 |
|
3325 |
if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
|
3326 |
iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; |
3327 |
} else if (bs->io_limits.iops[is_write]) { |
3328 |
iops_limit = bs->io_limits.iops[is_write]; |
3329 |
} else {
|
3330 |
if (wait) {
|
3331 |
*wait = 0;
|
3332 |
} |
3333 |
|
3334 |
return false; |
3335 |
} |
3336 |
|
3337 |
slice_time = bs->slice_end - bs->slice_start; |
3338 |
slice_time /= (NANOSECONDS_PER_SECOND); |
3339 |
ios_limit = iops_limit * slice_time; |
3340 |
ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write]; |
3341 |
if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
|
3342 |
ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write]; |
3343 |
} |
3344 |
|
3345 |
if (ios_base + 1 <= ios_limit) { |
3346 |
if (wait) {
|
3347 |
*wait = 0;
|
3348 |
} |
3349 |
|
3350 |
return false; |
3351 |
} |
3352 |
|
3353 |
/* Calc approx time to dispatch */
|
3354 |
wait_time = (ios_base + 1) / iops_limit;
|
3355 |
if (wait_time > elapsed_time) {
|
3356 |
wait_time = wait_time - elapsed_time; |
3357 |
} else {
|
3358 |
wait_time = 0;
|
3359 |
} |
3360 |
|
3361 |
bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3362 |
bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
|
3363 |
if (wait) {
|
3364 |
*wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
|
3365 |
} |
3366 |
|
3367 |
return true; |
3368 |
} |
3369 |
|
3370 |
static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors, |
3371 |
bool is_write, int64_t *wait)
|
3372 |
{ |
3373 |
int64_t now, max_wait; |
3374 |
uint64_t bps_wait = 0, iops_wait = 0; |
3375 |
double elapsed_time;
|
3376 |
int bps_ret, iops_ret;
|
3377 |
|
3378 |
now = qemu_get_clock_ns(vm_clock); |
3379 |
if ((bs->slice_start < now)
|
3380 |
&& (bs->slice_end > now)) { |
3381 |
bs->slice_end = now + bs->slice_time; |
3382 |
} else {
|
3383 |
bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
|
3384 |
bs->slice_start = now; |
3385 |
bs->slice_end = now + bs->slice_time; |
3386 |
|
3387 |
bs->io_base.bytes[is_write] = bs->nr_bytes[is_write]; |
3388 |
bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write]; |
3389 |
|
3390 |
bs->io_base.ios[is_write] = bs->nr_ops[is_write]; |
3391 |
bs->io_base.ios[!is_write] = bs->nr_ops[!is_write]; |
3392 |
} |
3393 |
|
3394 |
elapsed_time = now - bs->slice_start; |
3395 |
elapsed_time /= (NANOSECONDS_PER_SECOND); |
3396 |
|
3397 |
bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors, |
3398 |
is_write, elapsed_time, &bps_wait); |
3399 |
iops_ret = bdrv_exceed_iops_limits(bs, is_write, |
3400 |
elapsed_time, &iops_wait); |
3401 |
if (bps_ret || iops_ret) {
|
3402 |
max_wait = bps_wait > iops_wait ? bps_wait : iops_wait; |
3403 |
if (wait) {
|
3404 |
*wait = max_wait; |
3405 |
} |
3406 |
|
3407 |
now = qemu_get_clock_ns(vm_clock); |
3408 |
if (bs->slice_end < now + max_wait) {
|
3409 |
bs->slice_end = now + max_wait; |
3410 |
} |
3411 |
|
3412 |
return true; |
3413 |
} |
3414 |
|
3415 |
if (wait) {
|
3416 |
*wait = 0;
|
3417 |
} |
3418 |
|
3419 |
return false; |
3420 |
} |
3421 |
|
3422 |
/**************************************************************/
|
3423 |
/* async block device emulation */
|
3424 |
|
3425 |
typedef struct BlockDriverAIOCBSync { |
3426 |
BlockDriverAIOCB common; |
3427 |
QEMUBH *bh; |
3428 |
int ret;
|
3429 |
/* vector translation state */
|
3430 |
QEMUIOVector *qiov; |
3431 |
uint8_t *bounce; |
3432 |
int is_write;
|
3433 |
} BlockDriverAIOCBSync; |
3434 |
|
3435 |
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb) |
3436 |
{ |
3437 |
BlockDriverAIOCBSync *acb = |
3438 |
container_of(blockacb, BlockDriverAIOCBSync, common); |
3439 |
qemu_bh_delete(acb->bh); |
3440 |
acb->bh = NULL;
|
3441 |
qemu_aio_release(acb); |
3442 |
} |
3443 |
|
3444 |
static AIOPool bdrv_em_aio_pool = {
|
3445 |
.aiocb_size = sizeof(BlockDriverAIOCBSync),
|
3446 |
.cancel = bdrv_aio_cancel_em, |
3447 |
}; |
3448 |
|
3449 |
static void bdrv_aio_bh_cb(void *opaque) |
3450 |
{ |
3451 |
BlockDriverAIOCBSync *acb = opaque; |
3452 |
|
3453 |
if (!acb->is_write)
|
3454 |
qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size); |
3455 |
qemu_vfree(acb->bounce); |
3456 |
acb->common.cb(acb->common.opaque, acb->ret); |
3457 |
qemu_bh_delete(acb->bh); |
3458 |
acb->bh = NULL;
|
3459 |
qemu_aio_release(acb); |
3460 |
} |
3461 |
|
3462 |
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
|
3463 |
int64_t sector_num, |
3464 |
QEMUIOVector *qiov, |
3465 |
int nb_sectors,
|
3466 |
BlockDriverCompletionFunc *cb, |
3467 |
void *opaque,
|
3468 |
int is_write)
|
3469 |
|
3470 |
{ |
3471 |
BlockDriverAIOCBSync *acb; |
3472 |
|
3473 |
acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque); |
3474 |
acb->is_write = is_write; |
3475 |
acb->qiov = qiov; |
3476 |
acb->bounce = qemu_blockalign(bs, qiov->size); |
3477 |
acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb); |
3478 |
|
3479 |
if (is_write) {
|
3480 |
qemu_iovec_to_buffer(acb->qiov, acb->bounce); |
3481 |
acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); |
3482 |
} else {
|
3483 |
acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); |
3484 |
} |
3485 |
|
3486 |
qemu_bh_schedule(acb->bh); |
3487 |
|
3488 |
return &acb->common;
|
3489 |
} |
3490 |
|
3491 |
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
|
3492 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
3493 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3494 |
{ |
3495 |
return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); |
3496 |
} |
3497 |
|
3498 |
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
|
3499 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
3500 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3501 |
{ |
3502 |
return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); |
3503 |
} |
3504 |
|
3505 |
|
3506 |
typedef struct BlockDriverAIOCBCoroutine { |
3507 |
BlockDriverAIOCB common; |
3508 |
BlockRequest req; |
3509 |
bool is_write;
|
3510 |
QEMUBH* bh; |
3511 |
} BlockDriverAIOCBCoroutine; |
3512 |
|
3513 |
static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb) |
3514 |
{ |
3515 |
qemu_aio_flush(); |
3516 |
} |
3517 |
|
3518 |
static AIOPool bdrv_em_co_aio_pool = {
|
3519 |
.aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
|
3520 |
.cancel = bdrv_aio_co_cancel_em, |
3521 |
}; |
3522 |
|
3523 |
static void bdrv_co_em_bh(void *opaque) |
3524 |
{ |
3525 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3526 |
|
3527 |
acb->common.cb(acb->common.opaque, acb->req.error); |
3528 |
qemu_bh_delete(acb->bh); |
3529 |
qemu_aio_release(acb); |
3530 |
} |
3531 |
|
3532 |
/* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
|
3533 |
static void coroutine_fn bdrv_co_do_rw(void *opaque) |
3534 |
{ |
3535 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3536 |
BlockDriverState *bs = acb->common.bs; |
3537 |
|
3538 |
if (!acb->is_write) {
|
3539 |
acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, |
3540 |
acb->req.nb_sectors, acb->req.qiov, 0);
|
3541 |
} else {
|
3542 |
acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, |
3543 |
acb->req.nb_sectors, acb->req.qiov, 0);
|
3544 |
} |
3545 |
|
3546 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3547 |
qemu_bh_schedule(acb->bh); |
3548 |
} |
3549 |
|
3550 |
static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
|
3551 |
int64_t sector_num, |
3552 |
QEMUIOVector *qiov, |
3553 |
int nb_sectors,
|
3554 |
BlockDriverCompletionFunc *cb, |
3555 |
void *opaque,
|
3556 |
bool is_write)
|
3557 |
{ |
3558 |
Coroutine *co; |
3559 |
BlockDriverAIOCBCoroutine *acb; |
3560 |
|
3561 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3562 |
acb->req.sector = sector_num; |
3563 |
acb->req.nb_sectors = nb_sectors; |
3564 |
acb->req.qiov = qiov; |
3565 |
acb->is_write = is_write; |
3566 |
|
3567 |
co = qemu_coroutine_create(bdrv_co_do_rw); |
3568 |
qemu_coroutine_enter(co, acb); |
3569 |
|
3570 |
return &acb->common;
|
3571 |
} |
3572 |
|
3573 |
static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) |
3574 |
{ |
3575 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3576 |
BlockDriverState *bs = acb->common.bs; |
3577 |
|
3578 |
acb->req.error = bdrv_co_flush(bs); |
3579 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3580 |
qemu_bh_schedule(acb->bh); |
3581 |
} |
3582 |
|
3583 |
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs, |
3584 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3585 |
{ |
3586 |
trace_bdrv_aio_flush(bs, opaque); |
3587 |
|
3588 |
Coroutine *co; |
3589 |
BlockDriverAIOCBCoroutine *acb; |
3590 |
|
3591 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3592 |
co = qemu_coroutine_create(bdrv_aio_flush_co_entry); |
3593 |
qemu_coroutine_enter(co, acb); |
3594 |
|
3595 |
return &acb->common;
|
3596 |
} |
3597 |
|
3598 |
static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) |
3599 |
{ |
3600 |
BlockDriverAIOCBCoroutine *acb = opaque; |
3601 |
BlockDriverState *bs = acb->common.bs; |
3602 |
|
3603 |
acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); |
3604 |
acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); |
3605 |
qemu_bh_schedule(acb->bh); |
3606 |
} |
3607 |
|
3608 |
BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs, |
3609 |
int64_t sector_num, int nb_sectors,
|
3610 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3611 |
{ |
3612 |
Coroutine *co; |
3613 |
BlockDriverAIOCBCoroutine *acb; |
3614 |
|
3615 |
trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); |
3616 |
|
3617 |
acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque); |
3618 |
acb->req.sector = sector_num; |
3619 |
acb->req.nb_sectors = nb_sectors; |
3620 |
co = qemu_coroutine_create(bdrv_aio_discard_co_entry); |
3621 |
qemu_coroutine_enter(co, acb); |
3622 |
|
3623 |
return &acb->common;
|
3624 |
} |
3625 |
|
3626 |
void bdrv_init(void) |
3627 |
{ |
3628 |
module_call_init(MODULE_INIT_BLOCK); |
3629 |
} |
3630 |
|
3631 |
void bdrv_init_with_whitelist(void) |
3632 |
{ |
3633 |
use_bdrv_whitelist = 1;
|
3634 |
bdrv_init(); |
3635 |
} |
3636 |
|
3637 |
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
|
3638 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3639 |
{ |
3640 |
BlockDriverAIOCB *acb; |
3641 |
|
3642 |
if (pool->free_aiocb) {
|
3643 |
acb = pool->free_aiocb; |
3644 |
pool->free_aiocb = acb->next; |
3645 |
} else {
|
3646 |
acb = g_malloc0(pool->aiocb_size); |
3647 |
acb->pool = pool; |
3648 |
} |
3649 |
acb->bs = bs; |
3650 |
acb->cb = cb; |
3651 |
acb->opaque = opaque; |
3652 |
return acb;
|
3653 |
} |
3654 |
|
3655 |
void qemu_aio_release(void *p) |
3656 |
{ |
3657 |
BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p; |
3658 |
AIOPool *pool = acb->pool; |
3659 |
acb->next = pool->free_aiocb; |
3660 |
pool->free_aiocb = acb; |
3661 |
} |
3662 |
|
3663 |
/**************************************************************/
|
3664 |
/* Coroutine block device emulation */
|
3665 |
|
3666 |
typedef struct CoroutineIOCompletion { |
3667 |
Coroutine *coroutine; |
3668 |
int ret;
|
3669 |
} CoroutineIOCompletion; |
3670 |
|
3671 |
static void bdrv_co_io_em_complete(void *opaque, int ret) |
3672 |
{ |
3673 |
CoroutineIOCompletion *co = opaque; |
3674 |
|
3675 |
co->ret = ret; |
3676 |
qemu_coroutine_enter(co->coroutine, NULL);
|
3677 |
} |
3678 |
|
3679 |
static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, |
3680 |
int nb_sectors, QEMUIOVector *iov,
|
3681 |
bool is_write)
|
3682 |
{ |
3683 |
CoroutineIOCompletion co = { |
3684 |
.coroutine = qemu_coroutine_self(), |
3685 |
}; |
3686 |
BlockDriverAIOCB *acb; |
3687 |
|
3688 |
if (is_write) {
|
3689 |
acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, |
3690 |
bdrv_co_io_em_complete, &co); |
3691 |
} else {
|
3692 |
acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, |
3693 |
bdrv_co_io_em_complete, &co); |
3694 |
} |
3695 |
|
3696 |
trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); |
3697 |
if (!acb) {
|
3698 |
return -EIO;
|
3699 |
} |
3700 |
qemu_coroutine_yield(); |
3701 |
|
3702 |
return co.ret;
|
3703 |
} |
3704 |
|
3705 |
static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, |
3706 |
int64_t sector_num, int nb_sectors,
|
3707 |
QEMUIOVector *iov) |
3708 |
{ |
3709 |
return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); |
3710 |
} |
3711 |
|
3712 |
static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, |
3713 |
int64_t sector_num, int nb_sectors,
|
3714 |
QEMUIOVector *iov) |
3715 |
{ |
3716 |
return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); |
3717 |
} |
3718 |
|
3719 |
static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
3720 |
{ |
3721 |
RwCo *rwco = opaque; |
3722 |
|
3723 |
rwco->ret = bdrv_co_flush(rwco->bs); |
3724 |
} |
3725 |
|
3726 |
int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
|
3727 |
{ |
3728 |
int ret;
|
3729 |
|
3730 |
if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
|
3731 |
return 0; |
3732 |
} |
3733 |
|
3734 |
/* Write back cached data to the OS even with cache=unsafe */
|
3735 |
if (bs->drv->bdrv_co_flush_to_os) {
|
3736 |
ret = bs->drv->bdrv_co_flush_to_os(bs); |
3737 |
if (ret < 0) { |
3738 |
return ret;
|
3739 |
} |
3740 |
} |
3741 |
|
3742 |
/* But don't actually force it to the disk with cache=unsafe */
|
3743 |
if (bs->open_flags & BDRV_O_NO_FLUSH) {
|
3744 |
return 0; |
3745 |
} |
3746 |
|
3747 |
if (bs->drv->bdrv_co_flush_to_disk) {
|
3748 |
ret = bs->drv->bdrv_co_flush_to_disk(bs); |
3749 |
} else if (bs->drv->bdrv_aio_flush) { |
3750 |
BlockDriverAIOCB *acb; |
3751 |
CoroutineIOCompletion co = { |
3752 |
.coroutine = qemu_coroutine_self(), |
3753 |
}; |
3754 |
|
3755 |
acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); |
3756 |
if (acb == NULL) { |
3757 |
ret = -EIO; |
3758 |
} else {
|
3759 |
qemu_coroutine_yield(); |
3760 |
ret = co.ret; |
3761 |
} |
3762 |
} else {
|
3763 |
/*
|
3764 |
* Some block drivers always operate in either writethrough or unsafe
|
3765 |
* mode and don't support bdrv_flush therefore. Usually qemu doesn't
|
3766 |
* know how the server works (because the behaviour is hardcoded or
|
3767 |
* depends on server-side configuration), so we can't ensure that
|
3768 |
* everything is safe on disk. Returning an error doesn't work because
|
3769 |
* that would break guests even if the server operates in writethrough
|
3770 |
* mode.
|
3771 |
*
|
3772 |
* Let's hope the user knows what he's doing.
|
3773 |
*/
|
3774 |
ret = 0;
|
3775 |
} |
3776 |
if (ret < 0) { |
3777 |
return ret;
|
3778 |
} |
3779 |
|
3780 |
/* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
|
3781 |
* in the case of cache=unsafe, so there are no useless flushes.
|
3782 |
*/
|
3783 |
return bdrv_co_flush(bs->file);
|
3784 |
} |
3785 |
|
3786 |
void bdrv_invalidate_cache(BlockDriverState *bs)
|
3787 |
{ |
3788 |
if (bs->drv && bs->drv->bdrv_invalidate_cache) {
|
3789 |
bs->drv->bdrv_invalidate_cache(bs); |
3790 |
} |
3791 |
} |
3792 |
|
3793 |
void bdrv_invalidate_cache_all(void) |
3794 |
{ |
3795 |
BlockDriverState *bs; |
3796 |
|
3797 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
3798 |
bdrv_invalidate_cache(bs); |
3799 |
} |
3800 |
} |
3801 |
|
3802 |
void bdrv_clear_incoming_migration_all(void) |
3803 |
{ |
3804 |
BlockDriverState *bs; |
3805 |
|
3806 |
QTAILQ_FOREACH(bs, &bdrv_states, list) { |
3807 |
bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING); |
3808 |
} |
3809 |
} |
3810 |
|
3811 |
int bdrv_flush(BlockDriverState *bs)
|
3812 |
{ |
3813 |
Coroutine *co; |
3814 |
RwCo rwco = { |
3815 |
.bs = bs, |
3816 |
.ret = NOT_DONE, |
3817 |
}; |
3818 |
|
3819 |
if (qemu_in_coroutine()) {
|
3820 |
/* Fast-path if already in coroutine context */
|
3821 |
bdrv_flush_co_entry(&rwco); |
3822 |
} else {
|
3823 |
co = qemu_coroutine_create(bdrv_flush_co_entry); |
3824 |
qemu_coroutine_enter(co, &rwco); |
3825 |
while (rwco.ret == NOT_DONE) {
|
3826 |
qemu_aio_wait(); |
3827 |
} |
3828 |
} |
3829 |
|
3830 |
return rwco.ret;
|
3831 |
} |
3832 |
|
3833 |
static void coroutine_fn bdrv_discard_co_entry(void *opaque) |
3834 |
{ |
3835 |
RwCo *rwco = opaque; |
3836 |
|
3837 |
rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); |
3838 |
} |
3839 |
|
3840 |
int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
|
3841 |
int nb_sectors)
|
3842 |
{ |
3843 |
if (!bs->drv) {
|
3844 |
return -ENOMEDIUM;
|
3845 |
} else if (bdrv_check_request(bs, sector_num, nb_sectors)) { |
3846 |
return -EIO;
|
3847 |
} else if (bs->read_only) { |
3848 |
return -EROFS;
|
3849 |
} else if (bs->drv->bdrv_co_discard) { |
3850 |
return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
|
3851 |
} else if (bs->drv->bdrv_aio_discard) { |
3852 |
BlockDriverAIOCB *acb; |
3853 |
CoroutineIOCompletion co = { |
3854 |
.coroutine = qemu_coroutine_self(), |
3855 |
}; |
3856 |
|
3857 |
acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, |
3858 |
bdrv_co_io_em_complete, &co); |
3859 |
if (acb == NULL) { |
3860 |
return -EIO;
|
3861 |
} else {
|
3862 |
qemu_coroutine_yield(); |
3863 |
return co.ret;
|
3864 |
} |
3865 |
} else {
|
3866 |
return 0; |
3867 |
} |
3868 |
} |
3869 |
|
3870 |
int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) |
3871 |
{ |
3872 |
Coroutine *co; |
3873 |
RwCo rwco = { |
3874 |
.bs = bs, |
3875 |
.sector_num = sector_num, |
3876 |
.nb_sectors = nb_sectors, |
3877 |
.ret = NOT_DONE, |
3878 |
}; |
3879 |
|
3880 |
if (qemu_in_coroutine()) {
|
3881 |
/* Fast-path if already in coroutine context */
|
3882 |
bdrv_discard_co_entry(&rwco); |
3883 |
} else {
|
3884 |
co = qemu_coroutine_create(bdrv_discard_co_entry); |
3885 |
qemu_coroutine_enter(co, &rwco); |
3886 |
while (rwco.ret == NOT_DONE) {
|
3887 |
qemu_aio_wait(); |
3888 |
} |
3889 |
} |
3890 |
|
3891 |
return rwco.ret;
|
3892 |
} |
3893 |
|
3894 |
/**************************************************************/
|
3895 |
/* removable device support */
|
3896 |
|
3897 |
/**
|
3898 |
* Return TRUE if the media is present
|
3899 |
*/
|
3900 |
int bdrv_is_inserted(BlockDriverState *bs)
|
3901 |
{ |
3902 |
BlockDriver *drv = bs->drv; |
3903 |
|
3904 |
if (!drv)
|
3905 |
return 0; |
3906 |
if (!drv->bdrv_is_inserted)
|
3907 |
return 1; |
3908 |
return drv->bdrv_is_inserted(bs);
|
3909 |
} |
3910 |
|
3911 |
/**
|
3912 |
* Return whether the media changed since the last call to this
|
3913 |
* function, or -ENOTSUP if we don't know. Most drivers don't know.
|
3914 |
*/
|
3915 |
int bdrv_media_changed(BlockDriverState *bs)
|
3916 |
{ |
3917 |
BlockDriver *drv = bs->drv; |
3918 |
|
3919 |
if (drv && drv->bdrv_media_changed) {
|
3920 |
return drv->bdrv_media_changed(bs);
|
3921 |
} |
3922 |
return -ENOTSUP;
|
3923 |
} |
3924 |
|
3925 |
/**
|
3926 |
* If eject_flag is TRUE, eject the media. Otherwise, close the tray
|
3927 |
*/
|
3928 |
void bdrv_eject(BlockDriverState *bs, bool eject_flag) |
3929 |
{ |
3930 |
BlockDriver *drv = bs->drv; |
3931 |
|
3932 |
if (drv && drv->bdrv_eject) {
|
3933 |
drv->bdrv_eject(bs, eject_flag); |
3934 |
} |
3935 |
|
3936 |
if (bs->device_name[0] != '\0') { |
3937 |
bdrv_emit_qmp_eject_event(bs, eject_flag); |
3938 |
} |
3939 |
} |
3940 |
|
3941 |
/**
|
3942 |
* Lock or unlock the media (if it is locked, the user won't be able
|
3943 |
* to eject it manually).
|
3944 |
*/
|
3945 |
void bdrv_lock_medium(BlockDriverState *bs, bool locked) |
3946 |
{ |
3947 |
BlockDriver *drv = bs->drv; |
3948 |
|
3949 |
trace_bdrv_lock_medium(bs, locked); |
3950 |
|
3951 |
if (drv && drv->bdrv_lock_medium) {
|
3952 |
drv->bdrv_lock_medium(bs, locked); |
3953 |
} |
3954 |
} |
3955 |
|
3956 |
/* needed for generic scsi interface */
|
3957 |
|
3958 |
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) |
3959 |
{ |
3960 |
BlockDriver *drv = bs->drv; |
3961 |
|
3962 |
if (drv && drv->bdrv_ioctl)
|
3963 |
return drv->bdrv_ioctl(bs, req, buf);
|
3964 |
return -ENOTSUP;
|
3965 |
} |
3966 |
|
3967 |
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, |
3968 |
unsigned long int req, void *buf, |
3969 |
BlockDriverCompletionFunc *cb, void *opaque)
|
3970 |
{ |
3971 |
BlockDriver *drv = bs->drv; |
3972 |
|
3973 |
if (drv && drv->bdrv_aio_ioctl)
|
3974 |
return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
|
3975 |
return NULL; |
3976 |
} |
3977 |
|
3978 |
void bdrv_set_buffer_alignment(BlockDriverState *bs, int align) |
3979 |
{ |
3980 |
bs->buffer_alignment = align; |
3981 |
} |
3982 |
|
3983 |
void *qemu_blockalign(BlockDriverState *bs, size_t size)
|
3984 |
{ |
3985 |
return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size); |
3986 |
} |
3987 |
|
3988 |
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable) |
3989 |
{ |
3990 |
int64_t bitmap_size; |
3991 |
|
3992 |
bs->dirty_count = 0;
|
3993 |
if (enable) {
|
3994 |
if (!bs->dirty_bitmap) {
|
3995 |
bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) + |
3996 |
BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
|
3997 |
bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG; |
3998 |
|
3999 |
bs->dirty_bitmap = g_new0(unsigned long, bitmap_size); |
4000 |
} |
4001 |
} else {
|
4002 |
if (bs->dirty_bitmap) {
|
4003 |
g_free(bs->dirty_bitmap); |
4004 |
bs->dirty_bitmap = NULL;
|
4005 |
} |
4006 |
} |
4007 |
} |
4008 |
|
4009 |
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
|
4010 |
{ |
4011 |
int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; |
4012 |
|
4013 |
if (bs->dirty_bitmap &&
|
4014 |
(sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) { |
4015 |
return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] & |
4016 |
(1UL << (chunk % (sizeof(unsigned long) * 8)))); |
4017 |
} else {
|
4018 |
return 0; |
4019 |
} |
4020 |
} |
4021 |
|
4022 |
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
|
4023 |
int nr_sectors)
|
4024 |
{ |
4025 |
set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
|
4026 |
} |
4027 |
|
4028 |
int64_t bdrv_get_dirty_count(BlockDriverState *bs) |
4029 |
{ |
4030 |
return bs->dirty_count;
|
4031 |
} |
4032 |
|
4033 |
void bdrv_set_in_use(BlockDriverState *bs, int in_use) |
4034 |
{ |
4035 |
assert(bs->in_use != in_use); |
4036 |
bs->in_use = in_use; |
4037 |
} |
4038 |
|
4039 |
int bdrv_in_use(BlockDriverState *bs)
|
4040 |
{ |
4041 |
return bs->in_use;
|
4042 |
} |
4043 |
|
4044 |
void bdrv_iostatus_enable(BlockDriverState *bs)
|
4045 |
{ |
4046 |
bs->iostatus_enabled = true;
|
4047 |
bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; |
4048 |
} |
4049 |
|
4050 |
/* The I/O status is only enabled if the drive explicitly
|
4051 |
* enables it _and_ the VM is configured to stop on errors */
|
4052 |
bool bdrv_iostatus_is_enabled(const BlockDriverState *bs) |
4053 |
{ |
4054 |
return (bs->iostatus_enabled &&
|
4055 |
(bs->on_write_error == BLOCK_ERR_STOP_ENOSPC || |
4056 |
bs->on_write_error == BLOCK_ERR_STOP_ANY || |
4057 |
bs->on_read_error == BLOCK_ERR_STOP_ANY)); |
4058 |
} |
4059 |
|
4060 |
void bdrv_iostatus_disable(BlockDriverState *bs)
|
4061 |
{ |
4062 |
bs->iostatus_enabled = false;
|
4063 |
} |
4064 |
|
4065 |
void bdrv_iostatus_reset(BlockDriverState *bs)
|
4066 |
{ |
4067 |
if (bdrv_iostatus_is_enabled(bs)) {
|
4068 |
bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; |
4069 |
} |
4070 |
} |
4071 |
|
4072 |
/* XXX: Today this is set by device models because it makes the implementation
|
4073 |
quite simple. However, the block layer knows about the error, so it's
|
4074 |
possible to implement this without device models being involved */
|
4075 |
void bdrv_iostatus_set_err(BlockDriverState *bs, int error) |
4076 |
{ |
4077 |
if (bdrv_iostatus_is_enabled(bs) &&
|
4078 |
bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
4079 |
assert(error >= 0);
|
4080 |
bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : |
4081 |
BLOCK_DEVICE_IO_STATUS_FAILED; |
4082 |
} |
4083 |
} |
4084 |
|
4085 |
void
|
4086 |
bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes, |
4087 |
enum BlockAcctType type)
|
4088 |
{ |
4089 |
assert(type < BDRV_MAX_IOTYPE); |
4090 |
|
4091 |
cookie->bytes = bytes; |
4092 |
cookie->start_time_ns = get_clock(); |
4093 |
cookie->type = type; |
4094 |
} |
4095 |
|
4096 |
void
|
4097 |
bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie) |
4098 |
{ |
4099 |
assert(cookie->type < BDRV_MAX_IOTYPE); |
4100 |
|
4101 |
bs->nr_bytes[cookie->type] += cookie->bytes; |
4102 |
bs->nr_ops[cookie->type]++; |
4103 |
bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns; |
4104 |
} |
4105 |
|
4106 |
int bdrv_img_create(const char *filename, const char *fmt, |
4107 |
const char *base_filename, const char *base_fmt, |
4108 |
char *options, uint64_t img_size, int flags) |
4109 |
{ |
4110 |
QEMUOptionParameter *param = NULL, *create_options = NULL; |
4111 |
QEMUOptionParameter *backing_fmt, *backing_file, *size; |
4112 |
BlockDriverState *bs = NULL;
|
4113 |
BlockDriver *drv, *proto_drv; |
4114 |
BlockDriver *backing_drv = NULL;
|
4115 |
int ret = 0; |
4116 |
|
4117 |
/* Find driver and parse its options */
|
4118 |
drv = bdrv_find_format(fmt); |
4119 |
if (!drv) {
|
4120 |
error_report("Unknown file format '%s'", fmt);
|
4121 |
ret = -EINVAL; |
4122 |
goto out;
|
4123 |
} |
4124 |
|
4125 |
proto_drv = bdrv_find_protocol(filename); |
4126 |
if (!proto_drv) {
|
4127 |
error_report("Unknown protocol '%s'", filename);
|
4128 |
ret = -EINVAL; |
4129 |
goto out;
|
4130 |
} |
4131 |
|
4132 |
create_options = append_option_parameters(create_options, |
4133 |
drv->create_options); |
4134 |
create_options = append_option_parameters(create_options, |
4135 |
proto_drv->create_options); |
4136 |
|
4137 |
/* Create parameter list with default values */
|
4138 |
param = parse_option_parameters("", create_options, param);
|
4139 |
|
4140 |
set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size); |
4141 |
|
4142 |
/* Parse -o options */
|
4143 |
if (options) {
|
4144 |
param = parse_option_parameters(options, create_options, param); |
4145 |
if (param == NULL) { |
4146 |
error_report("Invalid options for file format '%s'.", fmt);
|
4147 |
ret = -EINVAL; |
4148 |
goto out;
|
4149 |
} |
4150 |
} |
4151 |
|
4152 |
if (base_filename) {
|
4153 |
if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
|
4154 |
base_filename)) { |
4155 |
error_report("Backing file not supported for file format '%s'",
|
4156 |
fmt); |
4157 |
ret = -EINVAL; |
4158 |
goto out;
|
4159 |
} |
4160 |
} |
4161 |
|
4162 |
if (base_fmt) {
|
4163 |
if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
|
4164 |
error_report("Backing file format not supported for file "
|
4165 |
"format '%s'", fmt);
|
4166 |
ret = -EINVAL; |
4167 |
goto out;
|
4168 |
} |
4169 |
} |
4170 |
|
4171 |
backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE); |
4172 |
if (backing_file && backing_file->value.s) {
|
4173 |
if (!strcmp(filename, backing_file->value.s)) {
|
4174 |
error_report("Error: Trying to create an image with the "
|
4175 |
"same filename as the backing file");
|
4176 |
ret = -EINVAL; |
4177 |
goto out;
|
4178 |
} |
4179 |
} |
4180 |
|
4181 |
backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT); |
4182 |
if (backing_fmt && backing_fmt->value.s) {
|
4183 |
backing_drv = bdrv_find_format(backing_fmt->value.s); |
4184 |
if (!backing_drv) {
|
4185 |
error_report("Unknown backing file format '%s'",
|
4186 |
backing_fmt->value.s); |
4187 |
ret = -EINVAL; |
4188 |
goto out;
|
4189 |
} |
4190 |
} |
4191 |
|
4192 |
// The size for the image must always be specified, with one exception:
|
4193 |
// If we are using a backing file, we can obtain the size from there
|
4194 |
size = get_option_parameter(param, BLOCK_OPT_SIZE); |
4195 |
if (size && size->value.n == -1) { |
4196 |
if (backing_file && backing_file->value.s) {
|
4197 |
uint64_t size; |
4198 |
char buf[32]; |
4199 |
int back_flags;
|
4200 |
|
4201 |
/* backing files always opened read-only */
|
4202 |
back_flags = |
4203 |
flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); |
4204 |
|
4205 |
bs = bdrv_new("");
|
4206 |
|
4207 |
ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv); |
4208 |
if (ret < 0) { |
4209 |
error_report("Could not open '%s'", backing_file->value.s);
|
4210 |
goto out;
|
4211 |
} |
4212 |
bdrv_get_geometry(bs, &size); |
4213 |
size *= 512;
|
4214 |
|
4215 |
snprintf(buf, sizeof(buf), "%" PRId64, size); |
4216 |
set_option_parameter(param, BLOCK_OPT_SIZE, buf); |
4217 |
} else {
|
4218 |
error_report("Image creation needs a size parameter");
|
4219 |
ret = -EINVAL; |
4220 |
goto out;
|
4221 |
} |
4222 |
} |
4223 |
|
4224 |
printf("Formatting '%s', fmt=%s ", filename, fmt);
|
4225 |
print_option_parameters(param); |
4226 |
puts("");
|
4227 |
|
4228 |
ret = bdrv_create(drv, filename, param); |
4229 |
|
4230 |
if (ret < 0) { |
4231 |
if (ret == -ENOTSUP) {
|
4232 |
error_report("Formatting or formatting option not supported for "
|
4233 |
"file format '%s'", fmt);
|
4234 |
} else if (ret == -EFBIG) { |
4235 |
error_report("The image size is too large for file format '%s'",
|
4236 |
fmt); |
4237 |
} else {
|
4238 |
error_report("%s: error while creating %s: %s", filename, fmt,
|
4239 |
strerror(-ret)); |
4240 |
} |
4241 |
} |
4242 |
|
4243 |
out:
|
4244 |
free_option_parameters(create_options); |
4245 |
free_option_parameters(param); |
4246 |
|
4247 |
if (bs) {
|
4248 |
bdrv_delete(bs); |
4249 |
} |
4250 |
|
4251 |
return ret;
|
4252 |
} |
4253 |
|
4254 |
void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs, |
4255 |
int64_t speed, BlockDriverCompletionFunc *cb, |
4256 |
void *opaque, Error **errp)
|
4257 |
{ |
4258 |
BlockJob *job; |
4259 |
|
4260 |
if (bs->job || bdrv_in_use(bs)) {
|
4261 |
error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs)); |
4262 |
return NULL; |
4263 |
} |
4264 |
bdrv_set_in_use(bs, 1);
|
4265 |
|
4266 |
job = g_malloc0(job_type->instance_size); |
4267 |
job->job_type = job_type; |
4268 |
job->bs = bs; |
4269 |
job->cb = cb; |
4270 |
job->opaque = opaque; |
4271 |
job->busy = true;
|
4272 |
bs->job = job; |
4273 |
|
4274 |
/* Only set speed when necessary to avoid NotSupported error */
|
4275 |
if (speed != 0) { |
4276 |
Error *local_err = NULL;
|
4277 |
|
4278 |
block_job_set_speed(job, speed, &local_err); |
4279 |
if (error_is_set(&local_err)) {
|
4280 |
bs->job = NULL;
|
4281 |
g_free(job); |
4282 |
bdrv_set_in_use(bs, 0);
|
4283 |
error_propagate(errp, local_err); |
4284 |
return NULL; |
4285 |
} |
4286 |
} |
4287 |
return job;
|
4288 |
} |
4289 |
|
4290 |
void block_job_complete(BlockJob *job, int ret) |
4291 |
{ |
4292 |
BlockDriverState *bs = job->bs; |
4293 |
|
4294 |
assert(bs->job == job); |
4295 |
job->cb(job->opaque, ret); |
4296 |
bs->job = NULL;
|
4297 |
g_free(job); |
4298 |
bdrv_set_in_use(bs, 0);
|
4299 |
} |
4300 |
|
4301 |
void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
|
4302 |
{ |
4303 |
Error *local_err = NULL;
|
4304 |
|
4305 |
if (!job->job_type->set_speed) {
|
4306 |
error_set(errp, QERR_NOT_SUPPORTED); |
4307 |
return;
|
4308 |
} |
4309 |
job->job_type->set_speed(job, speed, &local_err); |
4310 |
if (error_is_set(&local_err)) {
|
4311 |
error_propagate(errp, local_err); |
4312 |
return;
|
4313 |
} |
4314 |
|
4315 |
job->speed = speed; |
4316 |
} |
4317 |
|
4318 |
void block_job_cancel(BlockJob *job)
|
4319 |
{ |
4320 |
job->cancelled = true;
|
4321 |
if (job->co && !job->busy) {
|
4322 |
qemu_coroutine_enter(job->co, NULL);
|
4323 |
} |
4324 |
} |
4325 |
|
4326 |
bool block_job_is_cancelled(BlockJob *job)
|
4327 |
{ |
4328 |
return job->cancelled;
|
4329 |
} |
4330 |
|
4331 |
struct BlockCancelData {
|
4332 |
BlockJob *job; |
4333 |
BlockDriverCompletionFunc *cb; |
4334 |
void *opaque;
|
4335 |
bool cancelled;
|
4336 |
int ret;
|
4337 |
}; |
4338 |
|
4339 |
static void block_job_cancel_cb(void *opaque, int ret) |
4340 |
{ |
4341 |
struct BlockCancelData *data = opaque;
|
4342 |
|
4343 |
data->cancelled = block_job_is_cancelled(data->job); |
4344 |
data->ret = ret; |
4345 |
data->cb(data->opaque, ret); |
4346 |
} |
4347 |
|
4348 |
int block_job_cancel_sync(BlockJob *job)
|
4349 |
{ |
4350 |
struct BlockCancelData data;
|
4351 |
BlockDriverState *bs = job->bs; |
4352 |
|
4353 |
assert(bs->job == job); |
4354 |
|
4355 |
/* Set up our own callback to store the result and chain to
|
4356 |
* the original callback.
|
4357 |
*/
|
4358 |
data.job = job; |
4359 |
data.cb = job->cb; |
4360 |
data.opaque = job->opaque; |
4361 |
data.ret = -EINPROGRESS; |
4362 |
job->cb = block_job_cancel_cb; |
4363 |
job->opaque = &data; |
4364 |
block_job_cancel(job); |
4365 |
while (data.ret == -EINPROGRESS) {
|
4366 |
qemu_aio_wait(); |
4367 |
} |
4368 |
return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret; |
4369 |
} |
4370 |
|
4371 |
void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
|
4372 |
{ |
4373 |
/* Check cancellation *before* setting busy = false, too! */
|
4374 |
if (!block_job_is_cancelled(job)) {
|
4375 |
job->busy = false;
|
4376 |
co_sleep_ns(clock, ns); |
4377 |
job->busy = true;
|
4378 |
} |
4379 |
} |